A vertebrate-specific and essential role for osterix in osteogenesis revealed by gene knockout in the teleost medaka

Osteoblasts are bone matrix-forming cells that differentiate from Sox9- and Runx2-positive mesenchymal progenitors, which also give rise to chondrocytes and adipocytes (reviewed by Akiyama et al., 2005; Harada and Rodan, 2003; Ytteborg et al., 2015). The zinc-finger domain transcription factor Osx (also known as Sp7) is expressed in cells of the osteoblast lineage as well as in pre-hypertrophic chondrocytes, and is essential for embryonic and postnatal osteoblast differentiation in mice (Baek et al., 2010; Nakashima et al., 2002; Zhou et al., 2010). Severe bone defects caused by absence of osteoblasts, concomitant with ectopic cartilage formation, were reported in Osx null mice, suggesting that Osx is implicated in skeletal cell fate decisions (Koga et al., 2005; Nakashima et al., 2002). Importantly, a single base pair deletion in the OSX gene of a human patient with osteogenesis imperfecta has also been reported (Lapunzina et al., 2010). Therefore, a key regulatory role for Osx in mammalian bone formation is well-established, and many downstream targets and interactors have been identified (Hojo et al., 2016; Long, 2012). However, whether a similarly important role for bone formation is conserved also in non-mammalian vertebrates, including teleost fish, remains unknown. Two recent reports described the generation of osx mutants in members of the cyprinid family, the common carp (Cyprinus carpio) and zebrafish (Danio rerio) (Kague et al., 2016; Zhong et al., 2016). Interestingly, these mutants are viable and exhibit recognizable bone defects only at later stages of life. In contrast to observations in mouse, bone formation during embryonic and juvenile development was not blocked, raising doubts of a highly conserved role of osx during vertebrate osteogenesis. This suggested a delay in osteoblast maturation rather than a crucial function for osx in these cyprinid mutants.

Medaka is a teleost fish with intramembranous and perichondral formation of bone that, in contrast to mammals and zebrafish, lacks osteocytes (for an overview, see Witten and Huysseune, 2009). Medaka and zebrafish osx orthologues are expressed in pre-mature and mature osteoblasts, as in mice (DeLaurier et al., 2010; Renn and Winkler, 2009; Spoorendonk et al., 2008). In the present study, we generated medaka osx mutants and observed severe bone defects caused by defective bone matrix deposition and early lethality of mutant juveniles.

To trace the evolutionary origin and phylogenetic distribution of osx, we searched chordate genomes for homologues of medaka osx (Fig. 1A; see supplementary Materials and Methods for details). No such gene was found in the available genomes of tunicates, cephalochordates and lampreys. A clear orthologue of osx is present in the genomes of the elephant shark and the coelacanth. osx is also present in all teleosts and in spotted gar. Within tetrapods, we found homologues in amphibians, reptiles and mammals. A peculiar situation exists in birds. Although in most of the 71 available high quality avian genomes no trace of an osx homologue was found, in seven species an osx-like gene is present. Despite full conservation of the osx diagnostic DNA-binding domain (Hojo et al., 2016), higher divergence in the rest of the protein made inclusion of these genes in a phylogenetic tree inappropriate. Presence or absence of this gene does not follow any phylogenetic pattern. For confirmation, a conserved synteny analysis was done. osx was always found within a cluster of orthologous genes and tightly linked to its paralogue sp1 (Fig. S1). In bird species for which synteny information of the sp1 region is available, and in which we found an osx-like gene, it resides exactly at the expected position. In all other birds, there is no other gene or even traces of a gene from the Sp transcription factor family besides sp1. Extant jawless vertebrates (hagfish and lampreys) represent a group that never had a mineralized dermal skeleton (Janvier, 2015). It thus appears that osx emerged by a local genome duplication at the base of either the ‘ostracoderms’ (an informal term referring to jawless vertebrates with a dermal skeleton) or the jawed vertebrates (gnathostomes). osx was obviously lost repeatedly in the avian lineage and retained only in a handful of species. Further investigations are needed to explain how this loss could have occurred without disrupting bone formation.

osx Crispr_1 and Crispr_2 guide RNAs (gRNAs) were co-injected with Cas9 mRNA into osx:mCherry transgenic medaka embryos (Fig. 1B). The number and distribution of osx-positive osteoblasts were affected in the vertebral bodies of injected larvae (Fig. S2) suggesting osteoblast differentiation defects in this transient CRISPR (clustered regularly interspaced short palindromic repeats) assay. Only 23 out of 138 injected embryos developed into adulthood. Ten out of 23 did not show obvious skeletal or other malformations, and five of those were genotyped as mutant from fin clips (Fig. S2A). These were used for generation of F1 embryos. Different types of insertion-deletion (indel) mutations were detected in heterozygous F2 adult fish. Two alleles (osx^del138a, 25 bp deletion; osx^del137a, 28 bp deletion) result in frameshifts and premature stop codons at amino acid positions 138 and 137, respectively, upstream of the three zinc-finger DNA-binding domains (Fig. 1B). We performed qPCR analysis in osx mutants and found a 70% reduction in osx transcript levels (Fig. S3). We cannot exclude the possibility that the remaining transcripts contribute to residual Osx activity. However, this is unlikely given the positions of the introduced mutations, which are deduced to result in a truncated protein lacking the essential zinc-finger domain. Homozygous carriers for both alleles were obtained and showed identical phenotypes (Fig. S4). Therefore, only osx^del137a mutant phenotypes are described below.

Alizarin Red staining revealed reduction or absence of intramembranous bone formation in osx mutants, i.e. the operculum, parasphenoid and branchiostegal rays (Fig. 1C,E). Ultrastructural analysis showed reduced bone matrix formation in branchiostegal rays and cleithrum (Fig. S5). The number and patterning of teeth appeared normal in mutants, although tooth sizes were generally smaller in mutants compared with wild-type siblings (Fig. S6A,B). Detailed observations revealed that the smaller tooth size in mutants was due to less dentin being deposited and to the absence of attachment bone that normally connects the tooth to the underlying bone (Fig. S6C,D). Although the evolutionary history of attachment bone remains unresolved, this phenotype might be in line with a role for Osx in root, but not crown, dentin formation in mice (Zhang et al., 2015). The notochord sheath-based vertebral centra anlagen of osx mutants appeared to be normally mineralized (Fig. S7A,B). By contrast, the neural arches of vertebral bodies were completely lacking. No bone matrix and no Alizarin Red staining was detected (Fig. S7C,D).

Less than 1% of osx mutants (n>100 mutants analysed) survived beyond one month of age, and none beyond two months. Mutant one-month-old fish lacked scales and showed an almost complete absence of neural and hemal arches, as well as ribs, and lacked bony fin rays of all paired and unpaired fins (Fig. S8). They were also small in size and lacked most of the mineralized cranial skeleton. Thus, in this respect, medaka osx mutants recapitulate bone formation and mineralization defects reported in mouse mutants (Nakashima et al., 2002), confirming that osx is a crucial player in bone formation in non-mammalian vertebrates. However, unlike in the mouse model (Oh et al., 2012), the formation of cartilage appeared to be normal in medaka osx mutants (Fig. S9). Although the ectopic cartilage observed in mouse osx mutants (Oh et al., 2012) could be a consequence of delayed Osx-controlled perichondral ossification, we suggest that osx is dispensable for chondrocyte maturation in medaka.

Next, transgenic osteoblast reporter expression was analysed in wild-type and osx mutant larvae. Expression in the osx:mCherry reporter line closely follows the endogenous osx transcription pattern but is slightly delayed due to maturation of newly translated reporter protein (Renn and Winkler, 2009). Transgenic and endogenous osx expressions precede the onset of bone mineralization and continue in mature osteoblasts (Renn and Winkler, 2009). Expression of osteocalcin (osc) and the transgenic osc:eGFP reporter is found only in mature osteoblasts during ongoing ossification (Inohaya et al., 2007; Renn and Winkler, 2009). Also, collagen10a1 (col10a1) reporter expression in bone cells follows the endogenous expression pattern. However, col10a1:nuGFP reporter expression is also found in cartilage, where endogenous col10a1 expression cannot be detected at the corresponding stages, suggesting possible position defects at the transgene insertion site (Renn et al., 2013). Live imaging of osx:mCherry- and col10a1:nuGFP-labelled cells in the cranium of osx mutants (Fig. 1D,F) revealed an almost complete absence of col10a1-positive pre-osteoblasts in intramembranous bones such as the branchiostegal rays (Fig. 1F′) and operculum (Fig. 1F″).

Interestingly, although the number of osx:mCherry-positive cells was reduced in these bones, some cells persisted, which suggested that a limited number of pre-osteoblasts can form. These cells exhibit osx promoter activity but are unable to switch on other pre-osteoblast markers such as col10a1. Ultimately, these osx-deficient cells fail to differentiate into mature functional osteoblasts and are not able to produce collagen-containing osteoid. This was confirmed by analysing matured osteoblasts, which express osteocalcin (osc) (Inohaya et al., 2007). osc:eGFP-positive cells were completely missing in the operculum and branchiostegal rays of osx mutant fish (Fig. 1G,H). Over-exposure with 24-fold higher laser power revealed that only few cells with very low osc:eGFP expression remained in the cleithrum (Fig. S10), validating the conclusion that osteoblast maturation was generally blocked in osx mutants. This shows that osx is required for differentiation of most osteoblasts in medaka, with the exception of a few cells that start to differentiate into osteoblasts independently of osx, possibly due to an unknown compensatory mechanism.

Perichondral osteogenesis, i.e. bone formation on a previously established cartilage scaffold, was also affected in osx mutants. col10a1:nuGFP/osx:mCherry double-positive pre-osteoblasts were absent from the palatoquadrate (Fig. 2B′,F′). Interestingly, a few of these perichondral cells remained in the mutant ceratohyal (Fig. 2C-C″,G-G″), and these cells only expressed col10a1:nuGFP and not osx:mCherry. This might reflect a possible chondrocytic origin of these particular perichondral cells as proposed by Hammond and Schulte-Merker (2009) for zebrafish. These cells probably fail to fully differentiate into osteoblasts in the absence of osx function. Chondrocytes positive for col10a1:nuGFP appeared to form normally (Fig. 2B′,F′), consistent with our observation that cartilage matrix formation is normal in medaka osx mutants (Figs S4,S9).

As osteoblast differentiation was blocked in osx mutants, we next analysed whether osteoblast progenitor cells form normally. RNA in situ hybridization at 4 and 6 days post-fertilization (dpf) revealed that transcription of runx2, a general marker for osteoblast and chondrocyte progenitor formation (Flores et al., 2004; Li et al., 2009), was strongly reduced (Fig. 3A-D). By contrast, transcription of the chondrocyte marker col2a was normal, as expected (Fig. 3E,F). Comparable with our findings using the osx:mCherry reporter line, transcription of osx was slightly reduced especially in the parasphenoid and operculum (Fig. 3G,H), and it remains to be determined whether this reduction is due to nonsense-mediated decay of the mutant mRNA. Interestingly, col10a1 (Fig. 3I,J) and osc (Fig. 3K,L) transcripts were almost completely absent. This shows that not only is differentiation of osx cells into mature osc-expressing osteoblasts blocked in medaka osx mutants but the formation of runx2-positive osteoblast progenitors and col10a1-expressing pre-osteoblasts is also affected. These findings suggest that medaka osx has a so far unknown role in osteoblast progenitor specification. Alternatively, it is possible that formation of osteoblast progenitors depends on the presence of mature osteoblasts and/or mineralized bone. This is strikingly different from the situation in Osx null mice, in which Runx2 expression was not affected (Nakashima et al., 2002), but similar to the situation in zebrafish osx mutants at 6 weeks post-fertilization (Kague et al., 2016).

Besides ossification defects in the cranium, we also observed deficiencies in the developing vertebral column. Mineralized arches and their bone matrix were absent in osx mutants. By contrast, vertebral centra anlagen, which in teleost fish form by mineralization of the notochord sheath (Huxley, 1859; Kölliker, 1859), were present and regularly mineralized (Fig. 4A,F). Vertebral centra and arches constitute developmental modules (Lauder, 1980; Hautier et al., 2010). Recent studies did indeed confirm the independent development (modularity) of vertebral centra and arches. For example, in the stocksteif (dol, cyp26b1) mutant zebrafish and in malformed Atlantic salmon, vertebral centra fuse whereas arches remain separated (Witten et al., 2006; Laue et al., 2008; Spoorendonk et al., 2008). By contrast, zebrafish fused somites (fss, tbx6) mutants have fused arches but retain separated vertebral centra (van Eeden et al., 1996). Modularity of vertebral centra and arches is also displayed during normal development of teleosts, such as in the caudal fin endoskeleton of zebrafish (Bensimon-Brito et al., 2012a).

Live imaging showed that osx:mCherry- and osc:eGFP-positive cells were strongly reduced in number in both arches and centra (Fig. 4B,D,G,I). By contrast, the number of col10a1:nuGFP-positive cells was normal in the mutant vertebral centra (Fig. 4C,E,H,J). This supports the idea that col10a1:nuGFP-positive but osx:mCherry-negative cells regulate mineralization of the notochord sheath in medaka as suggested earlier (Renn et al., 2013). The strong similarities in composition between notochord sheath and cartilage matrix, and the fact that col10a1 expression is a prerequisite for cartilage mineralization (Kirsch and von der Mark, 1992) can further explain a requirement of col10a1:nuGFP-positive cells for notochord sheath mineralization. Future experiments analysing the activity of alkaline phosphatase in col10a1:nuGFP cells are needed to confirm the capacity of these cells to facilitate centra mineralization, similar to what has been shown for the notochord epithelium in salmon and zebrafish (Bensimon-Brito et al., 2012b; Grotmol et al., 2005).

Transmission electron microscopy revealed normal thickness of the notochord sheath (ns) in osx mutants and the almost complete absence of bone matrix outside the notochord (perichordal bone, pb) (Fig. 4K,L, black arrows), supporting the idea that osx-positive osteoblasts are not required for notochord sheath formation or for cartilage formation but are essential for bone matrix formation (Fleming et al., 2015).

In striking difference to recent reports on osx mutants in carp and zebrafish (Kague et al., 2016; Zhong et al., 2016), we observed strong defects in embryonic and larval bone formation after osx knockout in medaka, leading to early adult lethality in homozygous mutants and vertebral malformations in heterozygous carriers (Fig. S11). Generally, the bone defects in mutants appeared to be more severe than those that we previously reported in morphants after osx knockdown (Renn and Winkler, 2014). For example, col10a1 and osc expression was almost completely absent in the parasphenoid of mutants, whereas residual expression remained in morphants. The progenitor marker runx2 was strongly reduced in osx mutants but appeared normal in morphants (Fig. 3; compare with figure 4 in Renn and Winkler, 2014). Whereas bone defects, such as in the vertebral arches, eventually recovered in morphants, they persisted in mutants resulting in early lethality. Together, the mutant data provide evidence for an essential role of osx in osteoblast differentiation in medaka that was not completely evident in morphants.

Findings that osx-deficient carp and zebrafish show relatively normal initial bone patterning and formation could suggest that osx is dispensable for bone formation in these species. Given the role of osx in mammalian bone formation and given that it is required for bone formation in medaka, this explanation appears, however, unlikely. There are several possibilities to explain the incomplete phenotype of the osx mutants in carp and zebrafish. The carp has a tetraploid genome (Xu et al., 2014) and harbours two copies of sp7, both of which function in bone formation (Zhong et al., 2016). In such situations, the still-intact second copy can be expected to take over, at least partially, the function of the mutant copy. The sp7 gene of zebrafish has a peculiar genomic structure with a gain of two additional introns. The premature stop codon mutation lies in the small exon 2. A compensatory alternative splice event could skip this exon and create shorter transcripts that would code for a protein that still contains the majority of the wild-type protein, most importantly the DNA-binding zinc fingers. On a more general level, it is possible that other osx-related Sp genes, such as sp1, have a compensatory role in osteogenesis in zebrafish and carp. Sp7 and Sp1 – but also Sp3, and possibly other Sp factors – can bind to the same promoters, as has been shown in osteoblastic cells (Goto et al., 2006). Convincing evidence has also been presented that as-yet-unknown compensatory networks that involve genes unrelated to the one carrying the mutation become activated once a key developmental regulator is compromised (Rossi et al., 2015). It is conceivable that compensatory mechanisms are lineage specific and are in place in the two cyprinids (zebrafish and carp) but absent in medaka.

Still, our study highlights possible differences in aspects of osx function between mammals and teleosts. In medaka, either directly or indirectly, osx appears to control osteoblast progenitor specification, which has not been reported for mice. Despite these differences, our results demonstrate a significant degree of functional conservation between teleosts and mammals and suggest a key role for Osx in intramembranous and perichondral bone formation in non-mammalian vertebrates such as medaka.

All wild-type and transgenic medaka strains were described previously (Renn and Winkler, 2009; Renn et al., 2013). Fish were kept at 26°C under a controlled light cycle (14 h light, 10 h dark) to induce spawning. Embryos were kept in 0.3× Danieau's solution [19.3 mM NaCl, 0.23 mM KCl, 0.13 mM MgSO₄, 0.2 mM Ca(NO₃)₂, 1.7 mM HEPES, pH 7.0] at 30°C. All experiments were performed in accordance with approved Institutional Animal Care and Use Committee protocols of the National University of Singapore (R14-293). osx gRNAs were designed according to Hwang et al. (2013) and prepared as described in the supplementary Materials and Methods. Purified gRNAs (100 ng/μl each) were co-injected with Cas9 mRNA (300 ng/μl) into medaka embryos at the one-cell stage. Adult potential founders were genotyped by fin clipping and incrossed to obtain germline mosaic F1 embryos. Adult genotyped F1 fish were outcrossed with wild-type fish to obtain heterozygous F2 fish. Homozygous carriers in F3 were analysed for phenotype. For genotyping, larvae or adult fish were anaesthetized with 0.01% or 0.005% ethyl 3-aminobenzoate methanesulfonate (Tricaine; Sigma), respectively. Larvae or clipped caudal fin fragments were lysed individually in 50 μl of 50 mM NaOH and incubated at 95°C for 15 min. Samples were neutralized with 5 μl of 1 M Tris-HCl (pH 8.0). Stained embryos were fixed and washed with 1× phosphate buffered saline with 0.1% (v/v) Tween 20 (PBST) five times before genomic DNA isolation. Larvae were lysed individually with DNA lysis buffer (10 mM Tri-HCl pH 8.2, 50 mM KCl, 0.3% NP 40, 0.3% Tween) and Proteinase K (20 μg/ml; Sigma) with a ratio of 49:1 and incubated at 55°C for 60 min and 90°C for 10 min. The supernatant contained genomic DNA for analysis. A 430 bp fragment was amplified (using primers ACACTCGCTATGGCTCCAGT and GAGCCGTAAGGGTGTGTCAT) and digested with HaeIII (New England Biolabs). Mutants showed undigested fragments that were sequenced (Fig. S3).

Analyses of phylogenetic relationships and synteny conservation were carried out with sequences retrieved from publicly available genome data using previously published bioinformatic tools. For further details, see supplementary Materials and Methods.

In situ hybridization and Alcian Blue and Alizarin Red staining were carried out as described (Renn and Winkler, 2009). Microdissection of cartilage was performed according to DeLaurier et al. (2010). Imaging was performed according to To et al. (2012).

Specimens were fixed in a mixture of 1.5% glutaraldehyde/1.5% paraformaldehyde in 0.1 M cacodylate buffer for 24 h prior to decalcification in 10% EDTA for a minimum of 48 h. Specimens were postfixed with osmium tetroxide and embedded in Epon epoxy resin. Semi-thin sections (1 μm) were stained with Toluidine Blue for 1 min (0.2% Toluidine Blue, 2% Na₂CO₃), rinsed with water, air-dried, mounted with DPX (Fluka, Switzerland) and analysed with a Zeiss Axio Imager Z compound microscope. Ultrathin sections (0.02 μm) from the same blocks were contrasted with uranyl acetate and lead citrate and observed with a Jeol JEM 1010 transmission electron microscope operating at 60 kV. Images were digitized using a DITABIS drum scanner.

For live fluorescence imaging, larvae were anaesthetized with 0.01% Tricaine (Sigma) and pictures were taken using a Nikon SMZ1000 stereomicroscope equipped with NIS-Elements BR 3.0 software (Nikon). For live confocal imaging, larvae were anaesthetized with 0.005% Tricaine and embedded in 1.5% low melting agarose in a glass-bottom Petri dish. Confocal pictures were taken with a Zeiss LSM 510 Meta using 488 and 543 nm laser lines for GFP and mCherry analysis, respectively. Whole-mount fixed larvae and samples after microdissection were mounted in 100% glycerol for photography with a Nikon SMZ1000 stereomicroscope and a Nikon Eclipse 90i upright microscope equipped with NIS-Elements BR 3.0 software (Nikon). Imaging data were processed using Zeiss LSM Image Browser Version 4.2.0.121, ImageJ and Adobe Photoshop CS6 software.

Twenty larvae at 12 dpf were used for RNA extraction using the RNeasy Mini Kit (QIAGEN). Wild-type larvae served as control. All RNA samples were subjected to DNase I digestion. RNA was reverse transcribed using the RevertAid First Strand cDNA Synthesis Kit (Life Technologies). β-actin was used for normalization. The following primers were used: β-actin (TTCAACAGCCCTGCCATGTA, GCAGCTCATAGCTCTTCTCCAGGGAG); osx (TCTCCCCTCAGCTTCCTTAG, CTGGAAAGAGTGGGAGAAGG); osc (GAACCCGAGGTTATTGTGGA, TCACAGGCAACGTTCAGTTC).

cDNA of osx mutants and wild-type controls were used for qPCR. β-actin was used for normalization. The following primers were used: β-actin (GCCAACAGGGAGAAGATGAC, CATCACCAGAGTCCATGACG); osx (CAGATAAGACCGGCAGCAC, TCCTCCAGCTGTGGTGAAG). Comparisons of the gene expression levels in osx mutants relative to WT controls were performed in triplicate and analysed using Prism7000 software. A two-tailed Student's t-test was performed for statistical analysis.

We thank the members of the confocal unit of the NUS Centre for Bioimaging Sciences (CBIS) for their constant support.