The role of Pax-1 in axial skeleton development

The vertebral column is derived entirely from cells in the ventral halves of the somites. In morphological terms the events leading to axial skeleton formation have been extensively studied in several species, and are well described (Verbout, 1985; Christ and Wilting, 1992). The vertebral column is the most conspicuously segmented structure of the vertebrate body. The metameric pattern is initially laid down during the formation of somites, yet there is not a clear relationship between the vertebral units and the initial somites. Upon somite differentiation the ventral parts de-epithelialize and form the mesenchymal sclerotomes, which give rise to precartilaginous structures as well as connective tissue. This process is dependent on signals from the notochord (Watterson et al., 1954; Pourquie et al., 1993). Some sclerotome cells migrate medially and form the perichordal tube, which is initially unsegmented and uniform in density. Thereafter a segmented pattern of condensations starts to appear, the lateral regions slightly preceding the axial ones. The perichordal tube gives rise to vertebral bodies and intervertebral discs whereas the caudal halves of the paired lateral sclerotome areas are the origins of neural arches, pedicles and ribs (Verbout, 1985; Christ and Wilting, 1992). The axial regions of highest cell density, the intervertebral disc anlagen, are positioned approximately midsegmentally with respect to the former somite boundaries. By cell labelling experiments it has been shown that one vertebral body is derived from cells originating from two adjacent somites (Bagnall, 1992; Ewan and Everett, 1992). The mechanisms that control these morphological events in axial and lateral regions have not been studied, however.

The existence of a large number of mouse mutants with specific malformations in vertebral column development (Grüneberg, 1963; Theiler, 1988) is an important resource for investigating the molecular and morphogenetic events that lead to the development of this structure. One of these mutants, undulated (un; Wright, 1947), carries a point mutation in the Pax-1 gene (Balling et al., 1988), which is transcriptionally activated in sclerotome cells at the time of differentiation of the epithelial somites (Deutsch et al., 1988). A detailed morphological analysis of the skeletal phenotype and the embryonic development of this mutant has been made by Grüneberg (1950, 1954). He described malformations of vertebral bodies and intervertebral discs as well as vertebral processes, and came to the conclusion that these result from decreased sizes of the mesenchymal condensations that precede chondrification. Two alleles of un, undulatedextensive (un^ex ; Wallace, 1985) and Undulated short-tail (Un^s ; Blandova and Egorov, 1975), both being phenotypically more affected than un, have subsequently been isolated. Both these mutants have Pax-1 gene deletions, the latter one removing the entire Pax-1 locus (Balling et al., 1992; Dietrich S., Gruss P., and R.B., unpublished observations).

The Pax genes form a family of developmental control genes that have recently been the focus of great interest (Chalepakis et al., 1992; Gruss and Walther, 1992; Noll, 1993). They encode sequence-specific transcription factors that contain the DNAbinding paired-domain (Bopp et al., 1986; Chalepakis et al., 1991; Adams et al., 1992; Zannini et al., 1992), and in addition some of the Pax proteins also contain a paired-type homeodomain (Walther et al., 1991). In the mouse, nine Pax genes have been identified to date (Walther et al., 1991; Wallin et al., 1993), all being expressed in a spatially restricted manner beginning during early organogenesis. In addition to Pax-1, two other members of the Pax gene family have been shown to play important roles in mouse embryogenesis. Mutations in the Pax3 and Pax-6 genes cause the Splotch and Small-eye phenotypes, respectively (Epstein et al., 1991; Hill et al., 1991).

To gain further insight into early mechanisms of vertebral column formation we have carried out a comparative phenotypic analysis of the three Pax-1 mouse mutants, un, un^ex and Un^s, of which the latter two have not been investigated before. By analysis of mutant embryonic and new-born mice we show that Pax-1 is required for normal development of ventral vertebral structures, but is dispensable for sclerotome formation per se.

undulated (un) mice were purchased from the Jackson laboratory. Undulated-short tail (Un^s) and undulated-extensive (un^ex) mutant mice were kindly provided by Dr A. M. Malashenko, Krosnogorsk, Russia and Dr J. L. Cruickshank, Leeds, England, respectively. These mutants were backcrossed onto the C57BL/6 strain. On this genetic background homozygous un and un^ex mice show reduced fertility, particularly un^ex. Homozygous un^ex mice are often born small, in which case they also remain smaller and die earlier than their littermates. N1 to N5 generations were used for analysis. Embryos were recovered on days 9.5–18.5 p.c. where day 0.5 was 12 a.m. on the day of detection of the vaginal plug. On average, litters were born on day 19 p.c. As controls, +/+ or un/+ littermates were normally used. In exceptional cases C57BL/6 embryos of the corresponding stages were used.

Genotypings were made either by PCR analysis on embryonic DNA prepared from yolk-sacs or by Southern blotting of placental DNA. Genomic Southern blots were performed according to standard procedures (Sambrook et al., 1989). For PCRs two different primer pairs were used: (1) with primers flanking the mutated HaeIII restriction site in un and (2) specific for the fifth exon deleted in un^ex. These Pax1-specific oligonucleotides were:

1. 5′-CAGAGCAGACGTACGGCGAAG-3′ and 5′-AGGCAAA-GATGCCAGGATCCC-3′ and for

2. 5′-AGAGCCATCAGCATGGTTTCG-3′ and 5′-TGGAGGGAGTCCAGATTAAGC-3′. PCR reaction conditions were: 30 cycles of denaturation for 1 minute at 94°C, reannealing for 1 minute at 60°C and extension for 1 minute at 72°C for 1 minute; 2.5 units Taq polymerase (Amersham) using assay conditions specified by the manufacturer in a 50 μl reaction volume. PCR products were separated in agarose gels and visualized by ethidium bromide staining. For typing of un, PCR products were digested with HaeIII before gel separation, producing allele-specific band sizes that could easily be scored. For diagnosis of Un^s and un^ex homozygotes, a primer pair specific for an unlinked gene (Pax-9) was included as an internal control. A large number of un and un^ex embryos were also obtained from crosses of homozygous mice.

For histological analysis, day 9.5–18.5 embryos were fixed in Bouin’s solution and embedded in paraffin. 8 μm sections were stained either in haematoxylin-eosin or Azan. In addition, for day 10.5-12.5 p.c. embryos of the Un^s/Un^s, Un^s/un and +/+ genotypes, semi-thin sections were made, focusing on the strongly affected lumbar region. Embryos were fixed in Karnovsky’s solution, embedded in plastic and 0.75 μm sections were stained with toluidine blue.

Skeletal preparations were made by a slight modification of the Alcian blue/alizarin red staining procedure described by Kessel et al. (1990). Specimens were fixed in 99% ethanol for 24 hours (fetuses older than day 15 p.c. were first deskinned and eviscerated), and then kept in acetone for another 24 hours. Incubation in staining solution (1 vol. of 0.3% Alcian blue in 70% ethanol, 1 vol. of 0.1% alizarin red S in 96% ethanol, 1 volume of absolute acetic acid, and 17 volumes of 70% ethanol) was performed for 4–6 hours at 37°C and then overnight at room temperature. Samples were rinsed in water and kept in 1% potassium hydroxide/20% glycerol at 37°C over-night, with additional incubation at room temperature until complete clearing. For long term storage, specimens were transferred into 50%, 80% and finally 100% glycerol.

Pax-1 riboprobes were generated from a HincII-SacI paired box fragment using either ³⁵S-UTP or digoxigenin-11-UTP labelling for in situ hybridization on sections or in whole embryos, respectively. Fixation, hybridization and subsequent detection procedures were essentially as described by Kessel and Gruss (1991) and Rosen and Beddington (1993).

Cell proliferation was monitored by incorporation of BrdU into embryos in utero. Pregnant mice were injected intraperitoneally with a 5 mg/ml BrdU (Sigma) solution in PBS. The amount injected was approximately 50 μg/g body weight. After 2 hours the mice were killed and embryos isolated, fixed and embedded in paraffin. 8 μm sections were stained with an anti-BrdU antibody as described below and counterstained with basic fuchsin or hematoxylin. The number of labelled cells as well as total cell numbers were scored.

For immunostainings, embryos were fixed in 3% acetic acid in absolute ethanol at 4°C over-night, incubated first in absolute ethanol and then in xylene, both twice for 30 minutes, followed by a 1:1 mixture of xylene:paraffin for 30 minutes at 55°C, infiltrated with paraffin by three incubations for 1 hour at 55°C and embedded.

The generation of the Pax-1-specific antiserum has been previously described (Chalepakis et al., 1991). The mouse monoclonal anti-BrdU antibody was purchased from Dako.

For BrdU immunocytochemistry, sections were deparaffinized and air-dried over-night. They were then incubated in 2 N HCl for 30 minutes, washed in PBS, pH 6.0, for 1 minute and in PBS, pH 7.4 (hereafter called PBS) for 5 minutes. Primary antibody was diluted 1:50 in PBS, 1% BSA, 0.25% Tween-20 and incubated for 1 hour. Following washes, detection was made with the biotin-streptavidinperoxidase amplification procedure, UniTect ABC-kit (Dianova).

Pax-1 immunostaining was made on deparaffinized sections that were first bleached in 0.3% H₂O₂ in methanol for 30 minutes, washed in PBS, blocked with 10% normal goat serum, in PBS for 60 minutes and then incubated with the antiserum diluted 1:200 in the blocking solution. Detection was made with peroxidase-conjugated goat-antirabbit IgG (Sigma), and diaminobenzidine was used as chromogen.

To determine the role of Pax-1 in sclerotome differentiation, we have analysed Pax-1 RNA and protein expression in early somites and in subsequent axial skeleton development. Pax-1 transcription can first be detected around day 8.5 p.c. in the ventromedial part of newly formed somites (data not shown). This corresponds to the time of de-epithelialization of the ventral somite half, i.e. the emergence of the sclerotome. At first, expression is uniformly distributed in sclerotome cells. In whole-mount expression analysis of day-10.5 embryos with a Pax-1 RNA probe, it was possible to distinguish between cranial and caudal parts of the sclerotomes, the caudal part of each segment being more strongly positive and extending more dorsally than the cranial one (Fig. 1A). At day 12.5, the Pax-1 transcripts are mainly confined to the anlagen of the intervertebral disc and weaker domains can be seen in the perichondria lining the cartilage blastemas of the vertebral bodies, the pedicles and the proximal ribs (data not shown). These findings are in good accordance with those of Deutsch et al. (1988).

Pax-1 protein expression was detected with a rabbit antiserum raised against a peptide in the carboxy-terminal end of the protein (Chalepakis et al., 1991). The protein is expressed with a delay of about 1 day compared to the onset of transcription. First, at day 9.5, Pax-1 protein can be detected in sclerotome cells, albeit weakly (data not shown). Immunostaining of day-10.5 embryos revealed expression in almost all sclerotome cells (Fig. 1B). These cells continue to express Pax-1 up to the time when mesenchymal cells start to chondrify in order to form the vertebral bodies, which takes place around day 12.0 p.c. At day 12.5 p.c., the cells of the anlagen of the vertebral bodies are not stained, while the anlagen of the intervertebral discs remain strongly positive (Fig. 7E). Cells surrounding the vertebral body anlagen, those in the disc anlagen as well as those in the cranial half of each lateral sclerotome (forming connective tissue around the spinal nerves and dorsal root ganglia) remain strongly Pax-1-positive at day 14.5 p.c. (Figs 1C, 7G). In the vertebral column of new-born mice only a small number of Pax-1-positive cells can be observed in the ventral part of the annulus fibrosus (data not shown).

We have chosen to study the phenotype of three spontaneously arisen mouse mutants, un, un^ex and Un^s, since the molecular characterization of these mice has shown that there is a point mutation in the paired domain of un (Balling et al., 1988), whereas un^ex and Un^s display Pax-1 gene deletions. In Un^s the entire locus is deleted (schematically depicted in Fig. 2A). The absence of the Pax-1 gene in this mutant has previously been demonstrated for the paired box region (Wallin et al., 1993), and has been extended to the 5′-flanking sequences and 3′-untranslated region (data not shown). The size of the Un^s deletion has not been exactly determined, but a major chromosomal deletion has been excluded by karyotyping and by demonstration of the presence, in Un^s homozygous DNA, of several closely linked DNA markers (GBASE, March, 1993, A. Y. Hillyard, D. P. Doolittle, M. T. Davisson, and T. H. Roderick, The Jackson Laboratory, Bar Harbor, ME) (data not shown). In un^ex the fifth exon, which includes part of the carboxy-terminal coding sequence, is deleted (Dietrich S., Gruss P., and R. B., unpublished observations). The allelic nature of all three mutations has also been confirmed by inter-crosses. In no combination of the different mutants was a complementation effect observed.

The phenotypic appearance of adult mutant mice is characterized by their shortened and kinky tails (Fig. 2B-D). un is known as a recessive mutation (Wright, 1947), but in our colony we have observed a small percentage of heterozygous animals that have slight but significant distal tail kinks, which is also true for un^ex (data not shown). Thus although un/+ embryos occasionally do display very mild abnormalities, they serve as good controls for the stronger genotypes for all practical purposes. Un^s is invariably semidominant and produces a clear phenotype in the heterozygous situation. Un^s /Un^s mice die shortly after birth. Their tails are only rudimentary and the trunks considerably shorter than those of wild-type littermates (Fig. 3A,B), nevertheless pups show normal movements at birth. The reason for the postnatal lethality, which is also observed for the Un^s /un and Un^s /un^ex compound heterozygotes, could not be exactly determined. The most likely explanation is respiratory failure as a result of lung space constraints. Histological sections of Un^s /Un^s at day 18.5 p.c. show lungs with drastically reduced airway lumina (Fig. 3A,B).

Whereas the kinks in un/un are soft and can easily be corrected upon pressure, the tail abnormalities in un^ex /un^ex and Un^s /+ are more pronounced and rigid, partially due to vertebral fusions. The increase in severity of the phenotypes, can be interpreted as a gradual loss of Pax-1 function in the different mutants. In general, the axial skeleton abnormalities in these mutants are qualitatively similar but quantitatively different. This is more apparent when a detailed analysis of the vertebral components is made. As the strongest phenotypes result in postnatal death, we have concentrated our analyses and comparisons on embryos up to the new-born stage.

All undulated mutants display phenotypic abnormalities of the vertebral column along the entire axis. A list of typical vertebral column aberrations is given in Table 1. The abnormalities affecting vertebral bodies or other axial skeletal elements are similar in all the mutants, but can be found to different degrees; the Un^s /Un^s genotype being the most strongly affected. The severity of defects is not evenly distributed along the axis and is most pronounced in the lumbar region and the tail. A comparison of skeletal preparations at the new-born stage is shown in Fig. 4. In the wild type, single ossification centres are formed in all vertebral bodies (Fig. 4A). The un and un^ex homozygous new-borns show split vertebrae, vertebrae with dual ossification centers and missing transverse processes primarily in the lumbar region (Fig. 4B,C). These phenotypes are very similar, although un^ex has slightly more severe malformations and a few more segments are affected. In the case of Un^s a mild phenotype can already be seen in heterozygous mice (Fig. 4D). Vertebral body ossification centers are smaller than normal and transverse processes as well as the proximal parts of ribs are missing or reduced. Un^s /Un^s mice, with a total Pax-1 deficiency, completely lack vertebral bodies in the lower thoracic and lumbar regions (Fig. 4F). The compound heterozygote Un^s /un is intermediate in phenotype between Un^s /Un^s and un/un mice (Fig. 4E).

On closer inspection of the lower thoracic and upper lumbar regions, it is evident, that with the exception of Un^s /+, the different mutants lack intervertebral discs (Figs 3C,D and 5AE). In Un^s /+ the discs are thinner in the periphery, indicating a disturbance in the formation of the annulus fibrosus. The intervertebral disc is, at the new-born stage, composed partly of hyaline cartilage, and can therefore be stained with Alcian blue. In adult mice the hyaline cartilage has been replaced by fibrous cartilage (Theiler, 1988) and appears clear in wholemount skeletal preparations. In adult skeletal preparations the lack of discs in undulated could be confirmed (data not shown). In extreme cases the lack of discs leads to vertebral body fusions. On rare occasions mice of the Un^s /un^ex genotype have survived after birth and at most lived up to five months of age. The vertebral column of a few of these mice displayed multiple fusions (data not shown).

Thus, in all mutants we detect disturbed formation of both vertebral bodies and intervertebral discs. Furthermore, in all cases the proximal parts of the ribs as well as the rib homologues, the transverse processes, are affected. In un and un^ex the last pairs of ribs are shorter and fuse abnormally to the vertebral bodies. Un^s /+, Un^s /un and Un^s /Un^s display more severe abnormalities, with several floating rib pairs and the 13th pair is completely missing in Un^s /Un^s (Fig. 5A-E).

A systematic comparison of the +/+, un/un, Un^s /+, Un^s /un and Un^s /Un^s genotypes at the level of the ninth thoracic vertebra clearly shows that the basic phenotype is essentially the same, and varies only in degree of severity between the alleles (Fig. 5F-J). Although we observe small variations in phenotypic severity within one allele (none of the strains are completely inbred), these are insignificant compared to the variation between alleles.

Comparison of single vertebrae was made by dissection of whole-mount skeletal preparations. In Fig. 5K-T representative vertebrae from cervical, lumbar, sacral and caudal regions from wild-type and Un^s /Un^s mice are shown. From this analysis it is evident that vertebral body and rib/transverse process formation are abnormal at all levels. Where a ventral ossified structure can be seen, it resembles a pair of ventrally fused neural arches rather than a vertebral body (Fig. 5J,S). Therefore, Un^s /Un^s mice are deficient in structures derived both from the perichordal tube and the ventralmost parts of lateral sclerotomes. The neural arches are fairly normally shaped. Reductions of the spinous and articular processes as well as rare fusions between adjacent neural arches can be detected. These deformities are discrete in comparison to the ventral ones and may to a large extent be due to the compression of the vertebral column. The tail vertebrae display a phenotype which is different from the rest of axis. The tailremnant in Un^s /Un^s gives a trifurcated appearance, representing vestigial vertebral bodies and paired neural arches (Fig. 5T). The pedicles, the structures connecting bodies and neural arches, are lost. These malformed structures display only scant signs of segmentation and are fused to the tip of the tail.

To study when the Pax-1 protein starts to exert its effect we have conducted a detailed morphological analysis of mutant embryos. We have concentrated our study primarily on the lumbar region in the different mutants with particular emphasis on the strongest genotypes Un^s /un and Un^s /Un^s . The compound heterozygote Un^s /un has the advantage that the Pax-1 RNA and protein expressed from the un allele can be detected in the mutant embryos, although functionally this mutant is almost as affected as the null-allele Un^s /Un^s .

Analyses of cartilage formation in skeletal clearance preparations, revealed that axial structures in the vicinity of the notochord never chondrify in mutant embryos (Fig. 6B). The mutant phenotype is visible from day 11.5 p.c., the first timepoint when Alcian blue can be used to stain the developing skeleton. Therefore, in agreement with the earlier results of Grüneberg (Grüneberg, 1954), it can be concluded that the malformations are formed at an early embryonic stage and that the primary function of Pax-1 should be looked for before chondrogenesis commences. This lack of axial chondrogenesis most often leads to a lateral displacement of the notochord. Vertebral body formation is always more normal on the side of the notochord. This phenomenon, best observed in un^ex /un^ex skeletal preparations (Fig. 6C), suggests that vertebral body formation is also dependent on a late notochord function.

Histological analysis of early vertebral column formation, reveals that sclerotome formation occurs largely normally in the Pax-1-deficient mice. Morphological abnormalities can first be detected at day 10.5 p.c., when the lumbar sclerotome region already is abnormal compared with controls. In the wild type, densely and loosely arranged areas are present and the axial cells are oriented towards the notochord, whereas in the most affected mutants the sclerotome cells are evenly distributed (Fig. 7A,B).

At day 12.5 p.c. more profound changes can be detected. In mid-sagittal sections of Un^s /Un^s and Un^s /un embryos, in the region of the perichordal tube, cell numbers are strongly reduced, corresponding to the loss of the anlagen for the vertebral bodies and intervertebral discs (Fig. 7E,F). The few cells present in the perichordal tube of Un^s /un embryos at this stage do not display the normal metameric arrangement of condensed intervertebral disc anlagen and cell-sparse vertebral body anlagen. We performed immunostainings on mutant and wild-type embryos to compare the distribution of Pax-1expressing cells. As can be seen in Fig. 7, the Pax-1-positive domains are much wider in the craniocaudal dimension in the Un^s /un mutant. The observation that most cells are Pax-1-positive reflects the lack of chondrogenesis, which normally coincides with loss of Pax-1 expression as described previously. In Un^s /Un^s embryos, a more severe loss of cells in the perichordal tube could be observed and the total absence of Pax-1 protein could be confirmed (data not shown).

To analyse axial and lateral relationships, frontal sections of mutant embryos were made at day 10.5-12.5 p.c. In sections of the ventral part of the developing vertebral column in day-12.5 p.c. mutant embryos, a profoundly altered organization of the lateral lumbosacral sclerotome domains, in addition to the poorly developed perichordal tube, was revealed. Instead of the normal division in cranial and caudal halves, only one ‘compartment’ with equal, fairly high, cell density is detected. The spinal nerve is found in the middle, surrounded by concentric layers of fairly dense cells (Fig. 7C,D). These cells are all staining positive for Pax-1 (Fig. 7G,H). The effect is an expansion of mesenchymal regions at the expense of cartilage formation. Lack of the caudal condensations in these ventrolateral regions later results in the lack of transverse processes and proximal rib structures in the vertebral column of newborn mice. In more dorsal sections, through the anlagen of the neural arches, the craniocaudal division of each segment appears more normal. Thus, Pax-1 mutants do not have difficulty in establishing craniocaudal polarity per se, but rather display a loss of such polarity in the ventral domain of the sclerotome due to the loss of skeletal precursor structures.

When analysing sclerotome morphology in mutant embryos, we noticed that the notochord appeared larger than in wild-type controls at day 12.5–13.5 p.c. Transverse sections reveal more cells per section and a changed notochordal sheath in the mutants (Fig. 8). Consistent with this observation we found a dramatic increase in the rate of cell proliferation at the corresponding stage. In mutant embryonic notochord, a seven to tenfold increase in the number of BrdU-labelled nuclei was detected, at day 12.5 p.c. (Fig. 8E,F and Table 2). The finding that the development of the notochord is impaired is evident also at later stages. Whereas the normal notochord vanishes from the vertebral bodies and is supposed to contribute to the formation of the nucleus pulposus, the notochord in the strong Pax-1 mutants persists as a rod-like structure up to the new-born stage. This can be seen in skeletal preparations of the Un^s /un genotype most clearly (Fig. 5D).

Un^s homozygotes also display fused dorsal root ganglia (DRG) at day 18.5 p.c. (Fig. 3E). To some degree DRG fusions can be seen in all un alleles (data not shown). In early stages, up to around day 12.5 p.c., no fusions are detected. These arise later during fetal development, possibly as a consequence of the severe shortening of the trunk (Fig. 3B).

The proper formation of ventral vertebral structures requires Pax-1 function. Vertebral bodies and intervertebral discs are virtually absent in Pax-1-negative mice while the neural arches are fairly normal. This conclusion is obvious for the lumbar region of Un^s /Un^s mutants where ventral structures are completely missing, while other regions as the upper thoracic region present a more complex phenotype. Here the vertebrae have a ventral structure that connects the neural arches. Due to its form and small size we suggest that these are ventral neural arches and the perichordal tube contribution to this tissue, if any, is very small. Thus Pax-1-negative mice lack vertebral bodies at all levels of the vertebral column down to the beginning of the tail. The tail phenotype of Un^s /Un^s mice is different, however, due to the fact that the ventral cartilage remnants are not continuous with the paired lateral ones. The different phenotype in the tail may represent a different mechanism for vertebra formation by cells originating from the tail bud. When the perichordal tube has acquired a welldefined segmentation at about day 12.5 p.c., the intervertebral disc anlagen are Pax-1-positive whereas the vertebral body anlagen are negative. It may not be surprising, however, that both vertebral bodies and intervertebral discs are heavily affected in the mutants since the immunostaining data indicate that all or most cells that participate in the formation of ventral vertebral structures have expressed Pax-1 in their early developmental stages.

In Grüneberg’s original description of un, he stated that the intervertebral discs are larger than in normal mice (Grüneberg, 1954). This observation could not be confirmed in the present study. On the contrary, the annulus fibrosus is much reduced and cannot be observed in whole-mount skeletal preparations. The reason for this discrepancy is likely due to the fact that the anlagen of the intervertebral discs are less well defined and occupy a larger space in craniocaudal dimension in mutant embryos. These anlagen will not differentiate into proper discs, however. Our morphological observations and conclusions made from these, that are greatly facilitated by the allelic series now available, are otherwise in good agreement with Grüneberg’s work on un (1950, 1954).

The role of Pax-1 in intervertebral disc development might represent a late function compared to the patterning of early sclerotome cells. In later stages, Pax-1 expression is found in disc anlagen and lining the skeletal structures. Pax-1 may here have a role in the differentiation of fibrous tissue, and therefore, in the maintenance of the boundaries between skeletal elements. Indeed, fusions between axial skeletal elements are frequently seen in the mutants.

The strong phenotypic similarities in the three Pax-1 mutants leave little doubt that undulated is due to a reduction of Pax-1 function. While in un the phenotype can be linked to a single point mutation in the Pax-1 gene, the deletion sizes of the other two alleles are not known. It thus remains an open question whether other deleted genes also significantly contribute to the strong phenotype seen in Un^s /Un^s mice. We believe this not to be the case, because of the apparent lack of any phenotypic changes that cannot be observed in a milder form in un mice. This will not be strictly proven, however, until a clean Pax-1 null mutation has been accomplished via homologous recombination in ES cells and introduced into the germline or until Un^s /Un^s mice have been rescued via transgenesis. We suggest that Un^s /Un^s mice display a Pax-1 null phenotype and that the weaker mutations are hypomorphs. An observation in support of that idea is that Pax-1^un protein still retains DNA-binding activity, albeit much weaker than the wild-type protein (Chalepakis et al., 1991). Further support for this notion is provided by the demonstration that the causative genetic change in the Splotch-delayed mutant, is a point mutation in the paired domain, similar to the situation in un.; the phenotype of this mutant was interpreted to be due to partial loss of Pax-3 function (Vogan et al., 1993).

Recently, a detailed study of the interaction of paired domains with DNA target sequences revealed that this DNA binding domain is composed of two subdomains (Czerny et al., 1993). These authors discuss that a mutation in one of the subdomains may affect the binding to only a subset of target genes. This speculation is of great interest for the present study, as un is one of the potential models. Indeed, a differential effect on separate targets is consistent with our phenotypic analyses. Although all abnormalities to some degree can be seen in all mutants, the relative strength of phenotypic alterations in different regions are not always proportional. Thus, when the phenotype of un is compared to that of Un^s, it can be seen that the point-mutation in un affects development in axial sclerotome derivatives relatively more than lateral ones, i.e. the rib/transverse process phenotype is fairly weak in un.

We have recently identified a Pax gene that is highly homologous to Pax-1 (Wallin et al., 1993). This gene, Pax-9, is expressed in a similar but not identical pattern during sclerotome development. To analyse whether the expression of this gene is dependent on Pax-1 we have compared Pax-9 expression domains in wild-type and Un^s /un embryos (A. Neubüser, J. W. and R. B., unpublished observations). We have not been able to detect any changes in expression patterns that may not primarily be due to morphological changes.

A common denominator of the affected vertebral structures is their proximity to the notochord. The development of the vertebral column is dependent on the presence of the notochord (Watterson et al., 1954), and a supernumerary notochord induces the formation of additional vertebral body-like structures, but represses myotome formation, in surgically manipulated chicken embryos (Brand-Saberi et al., 1993; Pourquie et al., 1993). Interestingly, it was also shown that the notochord is able to induce Pax-1 expression in dorsal somite derivatives, that are normally Pax-1 negative (Brand-Saberi et al., 1993). The induced Pax-1 expression is accompanied by a loss of dermomyotome cells and the appearance of sclerotome-like mesenchymal cells. These observations make Pax-1 a very good candidate for a mediator of notochordal signals to the sclerotome, and as such might function as an embryonic competence factor. Based on the analysis of Pax-1 expression in the notochord mouse mutant Danforth’s short-tail (Sd) and comparison of the Sd and un phenotypes, we have suggested that Pax-1 might be one of the major mediators of inductive signals from the notochord to the sclerotome (Koseki et al., 1993). Our observation that vertebral body formation is often differentially affected on either side of the midline is correlated with a displacement of the notochord to the more normal side. We suggest that the development of the bodies is dependent on notochord signals that come significantly later than the factor(s) that initially induce Pax-1 expression. The simplest interpretation would be that sustained notochord signalling is required for normal maintenance of Pax-1 expression, which in turn controls the downstream cascade involved in the formation of vertebral bodies and intervertebral discs. Alternatively, Pax-1 acts as a competence factor in sclerotome cells enabling them to respond to further notochord signals.

A strong rib phenotype was observed in Un^s /Un^s . The proximal part is absent or reduced while lateral rib structures are unaffected. This division corresponds fairly well to those parts of the ribs that are apparently derived from medial and lateral somite-halves, respectively; the medial half gives rise to proximal ribs, whereas the lateral half gives rise to the more distal part of the rib (Charles Ordahl, personal communication). It may be that only the medial somite-half is dependent on notochord signals; in Sd, the rib phenotype is similar to the one in Un^s /Un^s (Koseki et al., 1993). Mutant mice with a phenotype that is virtually the opposite of that for Un^s with respect to the ribs have been recently described (Braun et al., 1992). These mice carry targeted mutations in the myf-5 gene.

To our surprise the morphological changes in the mutant also involve the notochord. Enlarged notochords were shown to have increased proliferation rates, and abnormal development results in the persistence of the notochord in vertebrae. These observations indicate that proper notochord development is dependent on surrounding sclerotome cells, which means that signalling may be bidirectional. Normally the notochord expands significantly at the level of the intervertebral discs, whereas it vanishes from the centres of the vertebral bodies. We suggest that notochordal cell proliferation is normally suppressed by surrounding perichordal tube cells and that subsequent differentiation of intervertebral discs leads to an expansion of the notochord only at these levels. A schematic model for the induction, expression and functional role of Pax-1 is presented in Fig. 9.

Another unexpected finding was the fusion of dorsal root ganglia. It is conceivable that this abnormality in a neural tissue is secondary to sclerotome abnormalities. It is well established that only the cranial half of each lateral sclerotome segment is permissive for neural crest cell migration and dorsal root ganglion growth (Keynes and Stern, 1984). In transplantation experiments in the avian system, it has been shown that neural crest-derived cell development in the context of cranial half somites only resulted in nonsegmented coalesced ganglia (Kalcheim and Teillet, 1989). In the Pax-1 mutants, the abnormal sclerotome cells in the cranial compartment may affect ganglion growth in a way akin to the influence on the notochord. Alternatively, the shortening of the vertebral column may cause a mechanical compression that forces the ganglia together, resulting in fusions.

From our observations in un mutant mice, we speculate that Pax-1 may act as an embryonic differentiation factor, making ventral mesenchymal cells competent to differentiate into hyaline and fibrous cartilage. Several mutant phenotypes resulting from genetic alterations in Pax genes have been described in diverse species (Chalepakis et al., 1992; Gruss and Walther, 1992; Noll, 1993), making the Pax gene family one of the currently better understood groups of developmental genes. Studies on the different Pax mouse mutants have documented a central role for Pax genes in morphogenesis. Splotch, a Pax-3 mutant, displays overgrowth of the neural tube with exencephaly and spina bifida, affecting also the migration of pigment cells, derivatives of the neural crest (Auerbach, 1954). Small-eye embryos, which carry a mutation in the Pax-6 gene, also display neural crest cell migration problems (Matsuo et al., 1993; Schmahl et al., 1993). Our data on the undulated phenotype might be interpreted as a problem of sclerotome cell proliferation, differentiation and/or migration. The significant reduction of ventral sclerotome cells has not been correlated to a readily detectable decrease in sclerotome cell proliferation rate (J. W. and R. B., unpublished observations). These preliminary data are, however, compatible with a slight but significant change in proliferation rate. Compared to the notochord, which displayed a large increase in proliferation rate, the developing sclerotome is a complex structure with regional differences in all three dimensions. An extended analysis is therefore required to document changes in the different sclerotome regions. One study that argues in favour of at least some proliferative role of Pax genes is the demonstration of an oncogenic activity of these genes (Maulbecker et al., 1993). Recently, a role for Pax-2 in the conversion of mesenchyme to epithelium during kidney development was demonstrated in an organ culture system (Rothenpieler and Dressler, 1993). This situation, where early mesenchymal cells fail to aggregate and differentiate, bears a strong resemblance to the phenotype in undulated mice where mesenchymal sclerotome cells fail to condense and initiate chondrogenesis. We propose that Pax-1 plays a crucial role in the early differentiation of ventral sclerotome cells, leading to profound changes in morphology and extracellular matrix composition. Very little is known about downstream target genes under control of Pax transcription factors. So far only three candidate target genes have been suggested; CD19 for Pax-5 and thyroperoxidase and thyroglobulin for Pax-8 (Kozmik et al., 1992; Zannini et al., 1992). The identification of Pax-1 as an essential gene for early sclerotome development provides a starting point, however, for a characterization of the cascade of events that underlie the morphogenesis of the axial skeleton.

We thank Peter Gruss for the gift of Pax-1 cDNA, Mary Dickinson and Andy McMahon for the BrdU-labelling protocol, Bärbel Strack, Monika Schüttoff, Günther Frank and Sybille Antoni for technical assistance, Uli Birsner for oligonucleotide synthesis and Lore Lay for photographic work. J. W. was supported by an EMBO fellowship and the Swedish Medical Research Council. H. K. was supported by the Human Frontier Science Program. The work was supported by the Max-Planck-Society and the Deutsche Forschungsgemeinschaft (Ba 869/3-1).