Specification of vertebral identity is coupled to Notch signalling and the segmentation clock

The vertebral column is a segmented structure characteristic for all vertebrates. The segmental pattern is established early during embryogenesis by the generation of somites. In most vertebrate species somites are blocks of epithelial cells that condense sequentially from the paraxial mesoderm on both sides of the neural tube in a strict anterior to posterior order. In mouse embryos, the first somites form in the posterior head-fold region at approximately day 7.75 of development. During the subsequent five days somite condensation progresses, while concomitantly new mesoderm cells are being generated caudally from the primitive streak and later from the tail bud elongating the embryo posteriorly (Gossler and Tam, 2002).

Somite formation and patterning require cell-to-cell communication in the presomitic mesoderm (psm) mediated by the Notch signalling pathway. Mutations in genes encoding Notch pathway components in mouse disrupt compartmentalization of somites, and alignment and maintenance of segment borders (Conlon et al., 1995; del Barco Barrantes et al.,1999; Evrard et al.,1998; Hrabé de Angelis et al., 1997; Kusumi et al.,1998; Swiatek et al.,1994; Zhang and Gridley,1998). Somite formation involves a molecular oscillator termed the`segmentation clock' that operates in the presomitic mesoderm and manifests itself by the periodic expression of `cyclic' genes. Expression of cyclic genes occurs in subsequent waves that sweep through the psm once during the formation of each somite (Forsberg et al.,1998; McGrew et al.,1998; Palmeirim et al.,1997). The segmentation clock is closely linked to Notch and Wnt/β-catenin signalling: cycling genes encode various Notch pathway components (Aulehla and Johnson,1999; Forsberg et al.,1998; Jiang et al.,2000; Jouve et al.,2000; McGrew et al.,1998; Palmeirim et al.,1997), and the negative regulator of the Wnt pathway axin2 (Aulehla et al.,2003). In addition, mutations in some Notch pathway components as well as in Wnt3a severely affect the expression of cyclic genes(Aulehla et al., 2003; Bessho et al., 2001; del Barco Barrantes et al.,1999; Jiang et al.,2000; Jouve et al.,2000).

During somite formation specific identities are imposed on somites according to their axial position (Gossler and Hrabe de Angelis, 1998; Hogan et al., 1985; Meinhardt, 1986). Transplantation experiments in chick(Kieny et al., 1972) and mouse(Beddington et al., 1992)embryos have indicated that positional information is established in the psm prior to the formation of epithelial somites. During subsequent somite differentiation, positional specification leads to unique morphologies of vertebrae along the body axis. Mutational analyses have shown that Hox genes are essential for the specification of vertebral identity(Krumlauf, 1994). During development Hox genes are activated sequentially according to their position in the cluster (Duboule,1994), leading to unique combinations of Hox genes expressed at different axial levels, which is referred to as `Hox code'(Kessel, 1991; Kessel and Gruss, 1991). In the paraxial mesoderm, Hox genes are generally activated in the posterior presomitic mesoderm and remain expressed in somites and their derivatives with distinct and appropriate expression boundaries. Recent analyses have shown that at least some Hox genes are additionally activated in bursts in the anterior psm, resulting in dynamic stripes that are correlated with the oscillating expression of cyclic genes(Zakany et al., 2001). In RBPjκ mutant embryos (Rbpsuh - Mouse Genome Informatics) expression of cyclic genes is disrupted(del Barco Barrantes et al.,1999) and Hox gene expression is reduced(Zakany et al., 2001). Likewise, loss or reduction of Wnt3a affects vertebral identity and expression of some Hox genes (Ikeya and Takada, 2001), suggesting a link between the segmentation clock,coordinated activation of Hox genes, and positional specification. However,thus far alterations of vertebral identities in mice with disrupted Notch signalling have not been reported.

To further study functions of Notch signalling in the paraxial mesoderm we have generated transgenic mice that express in the paraxial mesoderm a truncated version of Delta1 (Dll1 - Mouse Genome Informatics), Dll1^dn, which acts as a dominant-negative molecule in Xenopus and chick embryos (Chitnis et al., 1995; Henrique et al.,1997). These mice showed reduced Notch activity in the psm, were viable and displayed defects in somites and vertebrae consistent with known roles for Notch signalling in anteroposterior somite patterning. In addition, Dll1^dn transgenic mice showed with variable expressivity and penetrance alterations of vertebral identities consistent with homeotic transformations. Likewise, hemizygous transgenic Dll1^dnmice, which carried only one functional copy of the endogenous Dll1 gene, as well as mice heterozygous for the Dll1^lacZ null allele showed changes in vertebral identities, suggesting a previously unnoticed haploinsufficiency for Dll1. Also, in mice lacking Lfng function(Zhang and Gridley, 1998) or expressing Lfng constitutively in the presomitic mesoderm(Serth et al., 2003)identities of vertebrae were changed and axial identity was shifted anteriorly, indicating that levels of Notch signalling as well as cyclic activity of Lfng is essential for positional specification in the paraxial mesoderm.

The C-terminal deletion of 141 amino acids in Dll1 was generated by introduction of three stop codons and an XbaI site using PCR primers Dll1dn#1 (GTACCATGGGCCGTC) and Dll1dn#2(GGGTCTAGACTATTATCATTCAGGTGGAGGCTGGTG). The msd promoter is a 1494 bp FokI fragment fused to the Dll1 minimal promoter and exon one up to the ATG (Beckers et al.,2000). The Mesp2 promoter (a 1.2 kb PvuII/NcoIfragment 5′ of the ATG of the Mesp2 ORF) was subcloned from a Mesp2 clone isolated from a genomic λEMBL3a ES cell library(Schuster-Gossler et al.,1996). Both promoters were cloned in frame to the first ATG of the Dll1^dn ORF using NcoI sites. The Mesp2 promoter was additionally cloned in frame to the first ATG of lacZ. 3′to the coding regions a SV40 polyadenylation signal was included. Integrity of the constructs was verified by sequencing. Transgenic mice were generated by microinjection of linear construct DNA free of vector sequences into pronuclei of (BALB/cxC57BL6)F₁ fertilized eggs according to standard procedures. Dll1^lacz, msd::LfngHA3, and Lfng^lacZ mice were described previously(Hrabé de Angelis et al.,1997; Serth et al.,2003; Zhang and Gridley,1998).

PCR typing was performed using genomic DNA isolated from tail biopsies or yolk sacs, respectively. Used primers: msd::Dll1^dn and Mesp2::Dll1^dn: melta119 (CGGCTCTTCCCCTTGTTCTAAC) and Dll1-dn#5(TCTAGACTATTATCATTCAGGTGG); Mesp2::lacZ: lacZ3 (CAACTTAATCGCCTTGCAGC) and lacZ4 (CCAGATAACTGCCGTCACTCC); msd::LfngHA3: Lfng-F7 (CCTGTCCACTTTTGGTTTGC)and Lfng-B13 (CAGAGAATGGTCCCTTGATG); Lfng wild-type allele: lfhs1(GAACAAATATGGGCATTCACTCCA) and lfgwF13 (GGTCGCTTCTCGCCAGGGCGA); Lfng^lacZ allele: lfwF2 (CCAAGGCTAGCAGCCAATTAG) and lacZB2(GTGCTGCAAGGCGATTAAGTT); Dll1^lacZ allele: melta38(ATCCCTGGGTCTTTGAAGAAG) and lacZ1/Dll1ko (CAAATTCAGACGGCAAACG).

Dll1^dn transcripts in msd-Dll1^dn and Mesp2-Dll1^dn embryos were detected by RT-PCR on total RNA extracted from day 8.5-10.5 embryos using the RNeasy kit (Qiagen). Primers used were melta119 and Dll1-dn#5. HPRT expression was analysed as control using primers HPRT-5′ (CACAGGACTAGAACACCTGC) and HPRT-3′(GCTGGTGAAAAGGACCTCT).

In situ hybridisations on sections were performed according to Lescher et al. (Lescher et al., 1998). Whole-mount in situ hybridisations were performed according to Wilkinson(Wilkinson, 1992) with minor modifications using an InsituPro (Intavis AG number 10,000) for automated sample handling. Probes used were: Cdx1, Dll1, Lfng, Hes5, Hes7,myogenin, Pax1, Pax9, Nkx3.2, RalDH2, Tbx18, Uncx4.1, Hoxa4, Hoxa6, Hoxa9, Hoxb3, Hoxb4, Hoxb6, Hoxb8, Hoxc5, Hoxc6, Hoxc8,Hoxc9, Hoxd1, Hoxd3, Hoxd4 and Hoxd9. Hoxc5 and Hoxc6probes were generated from genomic PCR fragments using primers Hoxc5-1(ATGACTTTCTCACCTTCCTGCCCC), Hoxc5-2 (TCTCCTTCCCCAACACCTCTTTAC), Hoxc6-3(GTCATTTTGTCTGTCCTGGATTGG) and Hoxc6-4 (TCTGGATACTGGCTTTCTGGTCC). The SV40pA probe was generated from a 250 bp XbaI/BamHI subclone from the 3′ end of the msd-Dll1^dn construct.

Alcian blue/alizarin red skeletal staining was performed according to Kessel and Gruss (Kessel and Gruss,1991). Single vertebrae were dissected after staining of whole skeletons.

Culture of 9.5-day embryo tails was performed as described(Serth et al., 2003).

To analyse the consequences of reduced Notch signalling in the paraxial mesoderm we generated transgenic mice expressing a C-terminally truncated Dll1 cDNA, which was shown to act in a dominant-negative manner in Xenopusand chick embryos (Chitnis et al.,1995; Henrique et al.,1997). The mouse Dll1 cDNA was modified analogously such that only 12 amino acids of the intracellular portion adjacent to the transmembrane domain were retained (see Materials and methods). To express the truncated cDNA in the paraxial mesoderm we first used a region of the Dll1 promoter('msd') linked to the minimal promoter of Dll1, which was shown to direct gene expression specifically into the posterior presomitic mesoderm and newly formed somites of transgenic embryos(Beckers et al., 2000; Serth et al., 2003). The 3′ UTR and polyadenylation signal was derived from SV40(Fig. 1A) and served as RNA tag to confirm transgene expression in the paraxial mesoderm of transgenic embryos(Fig. 1B).

Three transgenic founder mice with conspicuous tail defects transmitted the transgene to their offspring, and transgenic lines designated msd::Dll1^dn13, 19 and 25 were established(Fig. 1C). Alcian blue/alizarin red-stained skeletal preparations showed that hemizygous mice of all three lines consistently displayed malformations of the vertebral column with incomplete penetrance and variable expressivity. Most prominent were fusions of neural arches predominantly in the cervical region, reduction or loss of pedicles and laminae in the cervical and thoracic regions, and split vertebral bodies (Fig. 1C, parts o-q, and data not shown).

If the truncated Dll1 acts also in mouse as a dominant-negative molecule and inhibits Notch signalling, increasing the dose of the transgene, or reducing endogenous Notch activity should enhance the observed defects. Consistent with this idea, external phenotypes and vertebral defects were more severe, and their penetrance was increased in homozygous transgenic mice from all three lines (n=7, 15 and 15, for msd::Dll1^dn13, 19 and 25, respectively (Fig. 1C,parts f-l, Fig. 2, and data not shown). Likewise in hemizygous msd::Dll1^dn19 mice that carried only one copy of the Dll1 WT allele, vertebral defects were enhanced(Fig. 1C, parts r-v, and Fig. 2D). In addition,expression of Hes5, which is a transcriptional target of Notch(Ohtsuka et al., 1999; Shimizu et al., 2002) and expressed in the psm of WT embryos in variable stripes(Fig. 3A,B, and data not shown), was not detected in the psm of homozygous msd::Dll1^dn19 embryos between day 9.5 and 11.5 of development, whereas expression in the neural tube was unaffected (Fig. 3C,D, and data not shown). Together, these data indicated that Dll1dn indeed acts in a dominant-negative manner and reduced Notch signalling in the psm.

The reduction or loss of pedicles in msd::Dll1^dn transgenic mice is reminiscent of, although milder than, the phenotype of mice lacking Uncx4.1 function (Leitges et al.,2000; Mansouri et al.,2000). Because Uncx4.1 expression is lost in the Dll1 null mutant (del Barco Barrantes et al.,1999), a reduction of Uncx4.1 could underlie the pedicle defects. Consistent with this idea, in homozygous transgenic day 9.5 msd::Dll1^dn19, and hemizygous msd::Dll1^dn19 day 9.5 embryos with only one copy of wild-type Dll1, Uncx4.1 expression in the prospective cervical and thoracic regions was reduced, similar to mouse embryos homozygous for a mutation that prevents efficient proteolytic processing of Notch1 (Huppert et al.,2000), and Tbx18 expression domains, which delineate anterior somite regions, were expanded(Fig. 3F,H,J,L, and data not shown). Pax9, which is normally expressed at high levels in the posterior-lateral sclerotome of each segment, showed less distinct domains of strong expression in posterior somite halves(Fig. 3N,P), whereas expression of other sclerotome markers such as Nkx3.2 and Pax1 was not obviously altered in msd::Dll1^dn embryos (data not shown). This suggested that somite compartmentalisation was affected in msd::Dll1^dn19 transgenic embryos, although distinct anterior and posterior compartments were clearly present. Analysis of myogenin, a marker for myotome, showed occasional subtle alterations of expression domains (see Fig. 8B), but fusions of adjacent myotomes that were found in Dll1 null embryos were not observed in transgenic day 9.5 and day 10.5 embryos (n=5, respectively).

In addition to the vertebral malformations described before, the ventral processes (anterior tuberculi) characteristic for the sixth cervical vertebra(C6) of WT mice (white arrowheads in Fig. 4A) were located abnormally in 74% of hemizygous msd::Dll1^dn19 mice (n=23)(Fig. 4B,B′, Table 1, and data not shown). Anterior tuberculi were either unilaterally or bilaterally reduced or missing on C6 (black arrowheads in Fig. 4B), which was accompanied by the unilateral presence at the seventh cervical vertebra (C7) of either a ventral structure or a transverse foramen (white arrowheads or arrow, respectively, in Fig. 4B), which is normally present at C3 to C6 (arrows in Fig. 4A) but not at C7. In addition, mice with a normal C6 and uni- or bilateral anterior tuberculi on C7 were observed(Table 1). Thus, in hemizygous msd::Dll1^dn19 transgenic mice C6 resembled at least in part C5, and C7 had acquired properties typical for C6, suggesting that the identities of C6 and C7 were altered. These alterations were enhanced and fully penetrant in animals of all three homozygous msd::Dll1^dn lines. In 40% of msd::Dll1^dn19 homozygotes (n=15) both ventral processes were missing at C6, and both sides of C7 carried either a ventral process or a transverse foramen (Fig. 4C,C′), whereas the remainder had unilateral transformations. In addition, in two cases the ribs attached to the first thoracic vertebra (T1) were reduced (grey arrowheads in lower-right panel in Fig. 4C), and in one such skeleton eight ribs were unilaterally attached to the sternum (data not shown). Overall, the observed alterations are consistent with anterior homeotic transformations in the cervical and upper thoracic regions. Reduction of endogenous Dll1 in hemizygous msd::Dll1^dn19 mice that carried one copy of the Dll1^lacZ null allele (n=16) led to alterations of vertebral identities in 87% of the cases(Fig. 4D,D′, Table 1). However, in contrast to hemizygous and homozygous msd::Dll1^dn19 mice, which exclusively displayed anterior transformations, phenotypes in some compound heterozygotes were more complex. In one case C6 appeared essentially normal, C5 carried small ventral processes that resembled small anterior tuberculi, and C7 had acquired ossified structures that might represent rudimentary ribs or portions thereof (left panels in Fig. 4D, and Table 1). In two other cases C6 had lost its characteristics and resembled C7, whereas C5 and C7 had acquired structures not typical for these vertebrae (right panels in Fig. 4D, and Table 1). Together, the alterations in these mice suggested that they carried composite anterior and posterior transformations. This prompted us to analyse in more detail the vertebral columns of heterozygous Dll1^lacZ mice(n=38). Four of these had rudimentary ribs uni- or bilaterally on C7(Fig. 4E,E′, Table 1, and data not shown),suggesting a previously unnoticed haploinsufficiency for Dll1 in the paraxial mesoderm. In contrast to msd::Dll1^dn19 mice, alterations in the cervical region resembled posterior transformations(Table 1).

The msd promoter fragment directed gene expression into the posterior psm and newly formed somites with a gap of expression in the anterior psm corresponding to S-I/S0 (Fig. 1B). To address whether the reduction of Notch signalling in the anterior psm is sufficient to induce alterations of vertebral identities, we generated transgenic mice carrying the truncated Dll1 cDNA under the control of the Mesp2 promoter (Fig. 5A). This directs heterologous gene expression in the anterior psm corresponding approximately to S-I(Haraguchi et al., 2001). Hemizygous transgenic mice of four independent lines carrying this construct were externally indistinguishable from non-transgenic littermates and had no obvious defects in their axial skeletons, although embryos expressed the transgene from day 9.5 onwards at levels readily detected by RT-PCR (lines #4 and #5 in Fig. 5C, and data not shown). Also homozygous Mesp2::Dll1^dn mice (homozygosity ascertained by test-matings with WT females or males, respectively) generated from line #4 were externally normal, although Notch activity was attenuated in the anterior psm as indicated by reduced Hes5 expression in homozygous day 9.5 (n=4) and 11.5 (n=2)Mesp2::Dll1^dn embryos (Fig. 5D). However, skeletal preparations showed changes in vertebral identities: four out of ten mice had only 5 lumbar vertebrae and 14 pairs of ribs with eight ribs attached to the sternum(Fig. 5E,F), whereas no malformations in the cervical region were found, which might be because of the absence of significant levels of transgene expression prior to day 9.5 (data not shown).

Numerous studies have shown that loss of individual Hox genes, alteration of their expression domains, or disruption of their temporal regulation causes homeotic transformations in mice (e.g. Jeannotte et al., 1993; Kessel et al., 1990; Kessel and Gruss, 1991; Ramirez-Solis et al., 1993; Zakany et al., 1997). To address whether Dll1^dn activity affects Hox gene expression, we analysed the expression of various Hox genes in homozygous msd::Dll^dn19 transgenic embryos between embryonic day 8.5 and 12.5. We focused on 15 genes whose published anterior expression borders lie in the regions with observed phenotypic alterations, and on genes that were shown by mutational analyses to be important for the specification of vertebrae, which were transformed in msd::Dll1^dn19 mice (for probes see Materials and methods). Whole-mount in situ hybridisation of transgenic day 8.5, 9.5 and 10.5 embryos did not show obvious spatial or temporal alterations of Hox gene expression in the paraxial mesoderm (data not shown). To analyse the spatial distribution of Hox gene transcripts more precisely, in situ hybridisations were performed with Hoxb3, Hoxb6, Hoxc5 and Hoxc6 on sagittal sections of paraffin embedded homozygous transgenic day 12.5 msd::Dll1^dn19 and with Hoxc8 on Mesp2::Dll1^dnembryos. In WT embryos (n=5) Hoxb6 expression was readily detected in prevertebra seven and subsequent posterior segments(Fig. 6A,A′) as described previously (Akasaka et al.,1996). However, in six out of 18 msd::Dll1^dn19 embryos Hoxb6 transcripts were only detected in and posterior to prevertebra eight (Fig. 6B,B′). Likewise Hoxc5 expression was barely detected in six out of 16 embryos in prevertebra C7 normally expressing the gene at high levels(Fig. 6E,E′), whereas the anterior expression borders of Hoxb3, Hoxc6 and Hoxc8 were unaltered (n=8, 7 and 7, data not shown). In addition, consistent with the low penetrance of posterior transformations, in two out of 12 heterozygous day 12.5 Dll1^lacZ embryos Hoxb6transcripts were detected ectopically in prevertebra six(Fig. 6C,C′', and data not shown). Interestingly, loss of Hoxb6 resulted in anterior transformations of C6 through T1 (Rancourt et al., 1995), suggesting that altered Hoxb6 expression contributes to the transformations in msd::Dll1^dn and Dll1^lacZ mice. To address whether the shift of Hoxb6 expression is evident earlier, we analysed day 8.5, 9.5 and 10.5 Dlldn embryos (n=15, 30 and 20, respectively) for Hoxb6expression by whole-mount in situ hybridisation. However, no consistent differences to WT embryos were detected. Homeotic transformations in the cervical and upper thoracic region were also found in mice lacking the caudal-type homeobox protein Cdx1(Subramanian et al., 1995),and reduced Cdx1 expression may cause transformations in Wnt3a mutants (Ikeya and Takada,2001) raising the possibility that Cdx1 expression might be affected in msd::Dll1^dn embryos. However, no alterations of Cdx1 expression were found in day 9.5 transgenic embryos (data not shown).

To address whether altered expression of cyclic genes in the psm might underlie altered vertebral identities and altered Hox gene expression in msd::Dll^dn embryos, we analysed expression of Lfng and Hes7, two cyclic genes essential for somite formation and patterning(Bessho et al., 2001; Evrard et al., 1998; Zhang and Gridley, 1998). Lfng patterns corresponding to all phases of the expression cycle were observed in transgenic day 9.5 embryos (n=39), with a distribution of embryos in the various phases of the expression cycle indistinguishable from WT embryos (n=40; data not shown). Embryo halves cultures also indicated cyclic Lfng expression in transgenic embryos (data not shown). Likewise, no obvious deviations from normal Hes7 expression (n=10) were observed, and Hoxd1expression in the anterior psm was indistinguishable from WT (n=13;data not shown). Thus, reduced Notch activity can change vertebral identities and alter Hox gene expression without readily detectable changes in cyclic gene expression in the psm.

Because several Hox genes are activated in transcriptional bursts that correlate with dynamic Lfng expression(Zakany et al., 2001), we analysed mice lacking Lfng function(Zhang and Gridley, 1998), and transgenic msd::LfngHA3 mice expressing Lfng constitutively in the psm (Serth et al., 2003). In embryos of both genotypes Hes5 expression was either downregulated or not detected (n=7, respectively, data not shown), indicating that both loss of Lfng function and constitutive Lfng activity affect Notch signalling similarly, and supporting that cyclic Lfngactivity is a critical parameter of Notch signalling in the psm. Vertebral malformations in these mice make it difficult to unambiguously assign axial identities to each vertebra. However, the ventral arch of C1, the anterior tuberculi normally present at C6, and the spinous process of T2 could be clearly distinguished and were used as landmarks to analyse this region of the axial skeleton. A consistent feature of all analysed skeletons of Lfng null mice (n=7) was a reduced number of cervical vertebrae and ribs (Fig. 7C). In five cases, the second vertebra anterior to the first rib-bearing vertebra T1 (C6 in WT mice) had two ventral processes resembling the anterior tuberculi present on WT C6. The remainder had one of these processes shifted to either the next anterior or posterior vertebra. In addition, in four skeletons the number of ribs attached to the sternum was reduced(Fig. 7). Transgenic mice with constitutive Lfng expression in the psm (n=11) showed similar defects. Five msd::LfngHA3 mice had only six cervical vertebrae,accompanied in part with additional alterations(Fig. 7). In all transgenic mice the number of ribs was reduced, and in four cases also the number of ribs attached to the sternum was altered (Fig. 7C).

The reduction of cervical vertebrae could reflect fusions of initially seven prospective cervical segments during sclerotome formation and subsequent differentiation, or could indicate an anterior shift of posterior axial identities. To distinguish between these possibilities we analysed the number of segments anterior to the fore limb bud in day 10.5 Lfng null or msd::LfngHA3 transgenic embryos well before formation of vertebral structures using myogenin expression as a marker for the segmentally arranged myotomes. In WT and homozygous msd::Dll1^dn19 embryos (n=4,respectively) myogenin expression indicated seven segments anterior to the fore limb bud (Fig. 8A,B). In contrast, two out of four hemizygous msd::LfngHA embryos had only six, and Lfng null embryos (n=5) had only five myogenin stripes anterior to the fore limb bud, respectively(Fig. 8C,D). Likewise, Dll1 null embryos (n=5), in which Lfng expression is severely downregulated, had only five segments preceding the forelimb bud(Fig. 8E). In addition, the number of segments between the fore and hind limb buds was reduced in Lfng null and hemizygous msd::LfngHA3 embryos: whereas 12-13 segments were present in WT (n=4) and homozygous msd::Dll1^dn19 embryos (n=4), only nine to 11 segments were found in Lfngnull (n=5) and msd::LfngHA3 (n=4) embryos (data not shown),consistent with the reduced number of thoracic vertebrae indicated by fewer ribs (Fig. 7). Because of fusions between adjacent myotomes in the trunk of homozygous Dll1^lacZ embryos(Hrabé de Angelis et al.,1997) the number of segments between fore and hind limb buds was difficult to assess unambiguously, but also appeared to be reduced. Together,these results suggested that the reduced number of cervical vertebrae was not because of sclerotome fusions, but the position of the limb buds, and thus axial identity is shifted anteriorly in embryos when cyclic Lfngexpression and thus probably cyclic Notch activity in the psm is abolished or severely downregulated. Consistent with this idea, the anterior-most expression of Hoxb6 was detected between prevertebrae four and five in sections of day 12.5 homozygous Lfng mutant embryos (n=2; Fig. 8G,G′) in contrast to prevertebra seven in WT embryos (Fig. 8F,F′).

We have shown that altered Notch signalling in the paraxial mesoderm of mouse embryos changes vertebral identities. Our results provide direct evidence that the coordination of segmentation and positional specification of mesoderm-derived tissues along the anteroposterior body axis requires both sufficient levels of Notch signalling as well as cyclic Lfngactivity.

The vertebral columns of msd::Dll1^dn transgenic mice showed mild defects, which probably reflect consequences of perturbed somite polarity. The largely normal vertebral column allowed us to also unambiguously identify the absence or ectopic presence of landmarks (e.g. anterior tuberculi and transverse foramina) characteristic for particular vertebrae. The loss of such landmarks from some vertebrae and their ectopic presence on others is indicative of altered vertebral identities and suggests homeotic transformations. Changes in vertebral identity were also found in mice expressing Dll1^dn under the Mesp2 promoter. In these mice no additional vertebral malformations were detected, suggesting that changes in vertebral identities in msd::Dll1^dn mice developed independently from segment polarity-related defects. The lack of vertebral defects indicative for disrupted anteroposterior somite patterning in Mesp2::Dll1^dn mice is surprising because Notch activity in the anterior psm is critically involved in somite compartmentalisation(Takahashi et al., 2003; Takahashi et al., 2000). A potential explanation might be that Notch activity in Mesp2::Dll1^dntransgenic embryos is higher than in msd::Dll1^dn embryos, as suggested by residual Hes5 expression(Fig. 5D), and still sufficient for establishment of segment polarity. The msd element directs mRNA expression into the posterior psm and newly formed somites but expression is weak or absent in the anterior region of the psm corresponding to S-I/S0(Beckers et al., 2000), whereas the Mesp2 promoter drives expression specifically in this region(Fig. 5B). Thus, in addition to apparently stronger reduction of Notch activity, in msd::Dll1^dnembryos paraxial mesoderm cells are exposed to reduced Notch activity during most of their progression through the psm, whereas in Mesp2::Dll1^dnembryos Notch activity is only reduced in cells shortly prior to somite formation, which might contribute to the phenotypic differences. Formally we cannot exclude that Dll1^dn expression in the newly formed somites contributed to changes in vertebral identity. However, this seems unlikely because expression in the anterior psm in Mesp2::Dll1^dn embryos was sufficient to affect vertebral identity. The apparent restriction of changes in vertebral identity to the cervical and upper thoracic region in msd::Dll1dn and heterozygous Dll1^lacZ mice might reflect a higher sensitivity of anterior Hox genes to reduced Notch activity.

The cervical region of transgenic mice expressing Dll1^dn showed anterior transformations, whereas heterozygous Dll1^lacZmice, which presumably have reduced Dll1-mediated signals, displayed posterior or bidirectional transformations, and Dll1^dn mice lacking one copy of Dll1 had mixed phenotypes. Dl1dn rendered chick retina cells deaf to receiving Notch signals (Henrique et al., 1997), blocking Notch activity cell autonomously. Thus,Dll1^dn appears to not only reduce Dll1-mediated Notch signals, but might also affect signals mediated by other ligands and various receptors, which could lead to different Notch signalling output in the psm than reduction of only Dll1. This might not only be a quantitative effect, because in both Dll1dn and Dll1dn/Dll1^lacZ/+embryos Hes5 expression was severely downregulated and not detected(Fig. 3 and data not shown). Such mode of action of the dominant-negative Dll1 could explain the different phenotypes in Dll1dn and Dll1dn/Dll1^lacZ/+ mice, and would imply that signals mediated by different ligands or receptors contribute to positional specification and might act potentially in opposite ways similar to non-redundant or even counteracting functions during somite compartmentalisation (Takahashi et al.,2003).

Based on the severe reduction of Hox gene expression in day 8.5 RBPjκ mutant embryos(Zakany et al., 2001), which lack Notch activity, one might expect that attenuated Notch signalling leads to reduced Hox expression. The results of the expression analysis of 15 Hox genes by in situ hybridisation in Dll1^dn embryos between day 8.5 and 10.5 did not provide evidence for this idea, although we cannot exclude subtle level differences that were not detected by our analysis. However, the anterior expression borders of Hoxb6 and Hoxc5 were shifted,suggesting that the exact positioning of the rostral limit of Hox expression requires precisely regulated Notch activity in the psm. General Hox activation in the paraxial mesoderm seems to be only affected significantly if Notch signalling is severely reduced or completely blocked potentially already in paraxial mesoderm precursors. How Notch activity and transcriptional regulation of Hox genes are coupled molecularly during paraxial mesoderm formation and patterning requires further investigation.

Transformations of vertebral identities, anterior shifts of Hoxb6expression, and of the position of both fore and hind limb buds were detected in mice lacking Lfng function or expressing Lfngconstitutively. An apparent anterior shift of Hoxb6 expression in Lfng mutant embryos could also be expected if fewer segments were generated in the prospective cervical region, whereas the absolute position of the anterior Hoxb6 expression border along the anteroposterior body axis was maintained. Recent models of somite segmentation suggest that the interaction of graded FGF (Dubrulle et al., 2001) or WNT (Aulehla et al., 2003) signals with the segmentation clock generates the periodic somite pattern. Conceptually, increasing the steepness of the gradient or slowing the periodicity of the clock would lead to fewer segments,which in either case would be larger. Thus, if the loss of Lfng would affect the clock (output) and fewer segments would be formed in the prospective cervical region, they should be larger than normal. However, the five cervical segments in Lfng mutant embryos occupied essentially the same space as the anterior five segments in WT embryos(Fig. 8), strongly supporting that the rostral Hoxb6 expression border is indeed shifted anteriorly. The positions of the fore and hind limb buds are invariant in WT embryos and correspond to the transition between the cervical and thoracic,and lumbar and sacral regions, respectively(Burke, 2000). Their anterior shift suggests homeotic transformations throughout the trunk region along the anterior posterior body axis that lead to an overall reduction of the number of segments in the trunk.

Experiments in chick embryos indicated that psm cells become determined with respect to the segmentation program and Hox gene expression in the anterior third of the psm at a level referred to as the `determination front',which appears to represent a threshold level of FGF8(Dubrulle et al., 2001). Extended exposure of cells to FGF8 in the anterior psm of chick embryos altered the position of somitic boundaries and shifted somitic Hox gene expression boundaries anteriorly (Dubrulle et al., 2001), and hypo- and hypermorphic mutations in FGFR1 caused homeotic transformations and subtle shifts of Hox gene expression borders in mouse embryos (Partanen et al.,1998), indicating that FGF signalling in the anterior psm has a critical role in positioning Hox expression borders. In mouse embryos transcriptional bursts of some Hox genes in the anterior psm correlated with cyclic Lfng expression, which has led to the idea that transcriptional regulation of Hox genes just prior to somite formation occurs as a response to the cyclic outcome of Notch activity, which might couple segmentation with the acquisition of axial identity(Zakany et al., 2001). Transformations in vertebral identities along the anteroposterior body axis in mice without Lfng function as well as with non-cyclic Lfngexpression in the psm provide direct experimental evidence that cyclic Lfng activity is essential to coordinate the generation of segments with their positional specification. Because the overall specification of different anatomical regions was maintained, colinearity of Hox gene expression was most probably not affected. It has been suggested(Dubrulle et al., 2001; Zakany et al., 2001) that assigning precise combinatorial Hox gene expression to somites occurs in two steps: first, most probably in precursors of the paraxial mesoderm prior to their entering the psm, Hox clusters would be progressively opened and become transcriptionally accessible. In the second step, in the anterior psm, the definitive expression of Hox genes would be allocated to segmental units coupled to the segmentation clock. Our results are consistent with a critical role of Notch signalling and cyclic Lfng activity in the second step,potentially after cells have passed the determination front. Thus, the interplay of FGF and Notch signals in the anterior psm might set definitive rostral Hox boundaries. Posterior transformations were achieved experimentally in transgenic mice by ectopic anterior expression of posterior Hox genes(Kessel et al., 1990; Lufkin et al., 1992; McLain et al., 1992). Likewise, loss of Lfng caused posterior transformations and an anterior shift of Hoxb6 expression. This suggests that Lfngactivity in the psm is required to prevent ectopic activation, or spreading of Hox gene expression anterior to their normal rostral expression boundaries. Thus, formally cyclic Lfng represses Hox gene transcription during setting the definitive anterior expression borders.

Taken together, our data demonstrate that both reduced Notch signalling without detected disruptions of cyclic gene expression in the psm as well as disrupted cyclic Lfng activity affect the positional specification of mesodermal derivatives along the anterior-posterior body axis. Thus, Notch signalling and most probably its cyclic modulation are required for the specification of vertebral identities.

We thank Jacqueline Deschamps, Denis Duboule, Mark Featherstone, Peter Gruss, Robb Krumlauf and Tom Lufkin for probes, and Jacqueline Deschamps for critical comments on the manuscript and discussion. This work was supported by the German Research Council (DFG SFB271).