Cdx1 and Cdx2 have overlapping functions in anteroposterior patterning and posterior axis elongation

caudal (cad) plays a role in anteroposterior (AP) patterning of the early fly embryo, and in specifying the identity of posteriormost structures. It encodes a protein of which the homeodomain shares approximately 50% amino acid sequence identity with the homeodomain of Antennapedia class proteins (Antp) and the two also partially share a conserved ‘hexapeptide’ motif upstream of the homeodomain. cad is initially expressed maternally. RNA is first transcribed ubiquitously but it is differentially degraded by the maternal Bicoid, to generate a posterior gradient. Later in development, the maternal RNA is replaced by zygotic transcripts, which also accumulate at the posterior end of the embryo (Mlodzik et al., 1985; Mlodzik and Gehring, 1987; MacDonald and Struhl, 1986). The maternal and the zygotic cad gradient are necessary for proper segmentation of the posterior blastoderm (MacDonald and Struhl, 1986). They control the expression of gap and pair-rule genes, which in turn regulate the homeotic (HOM-C) genes (Dearolf et al., 1989; Rivera-Pomar et al., 1995; Schulz and Tautz, 1995). After gastrulation, the zygotic expression domain of cad immediately flanks the posterior expression boundary of the most posteriorly expressed HOM-C gene, Abd-B. cad is necessary for the specification of the most posterior segment, the analia, and in this respect behaves as a HOM-C gene (Moreno and Morata, 1999). cad like genes have been found in a variety of species. The Amphioxus homologue AmphiCdx has been found to be linked with the homeobox genes AmphiXlox and AmphiGsx, in a recently defined ParaHox cluster, which is presumed to be an ancient paralogue of the Hox clusters. The Hox and ParaHox clusters have probably arisen by duplication of an original ‘ProtoHox’ cluster (Brooke et al., 1998). In the mouse, a remnant of a ParaHox cluster consists of Gsh1, Pdx1 and the cad homologue Cdx2 (Chawengsaksophak and Beck, 1996; Beck et al., 2000). Like the Hox genes, ParaHox genes display a conserved co-linear expression along the AP axis in neural tube, in mesoderm and in the gut endoderm (Beck et al., 1995; Guz et al., 1995; Li et al., 1996; Brooke et al., 1998). No link between the two other mouse cad homologues, Cdx1 and Cdx4, with potential paralogues of Pdx and Gsh on possible duplicates of the ParaHox cluster has been reported.

Cdx1 starts to be expressed at the late primitive streak stage (day 7.2) in ectodermal and mesodermal cells of the primitive streak (Meyer and Gruss, 1993). Expression of Cdx2 begins at day 3.5 in the trophectoderm of the blastocyst and later continues in tissues derived from the extra-embryonic ectoderm. Cdx2 transcripts are detected in the embryo proper slightly later than Cdx1 (Beck et al., 1995) (E. Tolner, B. Roelen and J. D., unpublished). Initiation of Cdx4 expression follows, in the allantois and the posterior part of the primitive streak in the late streak stage embryo (Gamer and Wright, 1993) (E. Tolner, B. Roelen and J. D., unpublished). The three Cdx genes, although paralogous, exhibit overlapping expression patterns in the neural tube and mesoderm at day 8.5 of development. Their expression domains in these tissues form nested sets, with Cdx1 showing most anterior and Cdx4 most posterior rostral expression boundaries, compatible with a possible differential role in AP patterning (Gamer and Wright, 1993; Meyer and Gruss, 1993; Beck et al., 1995). This has been confirmed by the generation of loss-of-function mutants. Cdx1 null mutant mice are viable and show anterior homeotic transformation of axial skeletal elements from the level of the first cervical (C1) to the level of the eighth thoracic vertebra (T8) (Subramanian et al., 1995). Cdx2 null mutant mice die at the time of implantation, but heterozygous mice are viable and display anterior homeosis of vertebrae from the level of C6 to the level of T8, and slightly shorter tails. Cdx2 and Cdx4 are expressed in posterior gut endoderm at day 8.5 (Gamer and Wright, 1993; Beck et al., 1995). Cdx2 expression has been shown to be maintained in the gut epithelium at later stages, and to persist during adulthood (James and Kazenwadel, 1991). Cdx1 starts to be expressed later (from day 14.5) in developing intestinal epithelium where it becomes restricted and is maintained in the proliferative crypt compartment (Duprey et al., 1988; Meyer and Gruss, 1993; Subramanian et al., 1998). Cdx2^+/– mice develop polypoid lesions in the intestine secondary to homeotic transformation of the distal intestine, showing that Cdx genes have a function in gut development (Chawengsaksophak et al., 1997; Beck et al., 1999).

Experiments in mouse and Xenopus suggest that Cdx genes may be directly involved in the regulation of Hox genes (Pownall et al., 1996; Epstein et al., 1997; Isaacs et al., 1998; Charité et al., 1998). Cdx1 null mice showed posterior shifts in the mesodermal expression boundary of a number of Hox genes (Subramanian et al., 1995). Cdx-binding sites are present in Hox regulatory sequences (Shashikant et al., 1995; Subramanian et al., 1995; Charité et al., 1998) and Cdx1 can transactivate Hoxa7 reporter constructs (Subramanian et al., 1995). We have previously shown that Cdx proteins in vitro bind Cdx-binding sites in a regulatory element of Hoxb8. These sites were shown to be necessary for the ability of this element to drive Hox-like expression in transgenic embryos. Expression of both transgene and endogenous Hoxb8 was anteriorised by rostral overexpression of Cdx genes in neurectoderm and mesoderm at day 8.5 (Charité et al., 1998).

In an attempt to elucidate the relationship between Cdx genes and the Hox genes, we have investigated the effect of loss-of-function mutations in Cdx genes on AP patterning and Hox gene expression. We show that Cdx1^–/–/Cdx2^+/– double mutants exhibit skeletal defects along the complete vertebral column, with an increased severity compared with single mutants. An associated posterior shift of Hox gene expression is also more extensive in these double mutants than in single mutants, demonstrating a parallel functional redundancy between Cdx genes in regulating Hox gene transcription. Furthermore, we show that mesodermal and neurectodermal expression boundaries of trunk Hox genes are already affected at day 8.5/9.0 in Cdx mutants, suggesting that Cdx genes are involved in modulating the early establishment phase of Hox expression at these levels. However, Cdx1 co-operates with Cdx2 in ensuring proper posterior axis elongation from the tail bud at later stages. We discuss the possibility that Cdx genes could, depending on the time and A-P position, influence early patterning by regulating Hox genes, and act as later mediators of posterior axial extension, directly or as regulators of 5′ Hox genes.

The generation of Cdx1 and Cdx2 loss-of-function mutants and of the Hoxb8lacZneo– ‘knock in’ line have been described previously (Subramanian et al., 1995; Chawengsaksophak et al., 1997; van den Akker et al., 1999). Cdx1 mutant mice were from a mixed background, and have been re-derived from crosses with C57Bl6/CBA F₁ hybrids. They were then bred with these F₁ hybrids for about six generations, and were subsequently made homozygous. The Cdx2 mutant mice underwent the same history, except that they continued to be bred as heterozygotes with C57Bl6/CBA F₁ mice. The Hoxb8 knock in strain was originally generated from 129 Ola ES cells, and the mutation has been kept on an FVB background after the CRE deletion step.

Mice heterozygote for the Hoxb8 null allele wherein lacZ is inserted in frame in the Hoxb8 locus are named Hoxb8lacZ^+/–. Cdx1-null mice were crossed with Cdx2^+/– animals to obtain transheterozygous offspring. These were mated with Cdx1-null mice and the progeny was used for skeletal staining. We also used these crosses to recover single and double mutant day 8.5 and 9.5 embryos for whole mount in situ hybridisation, and day 14.5 embryos for histological analysis on sections. For analysis of the Hoxb8 expression pattern in the Cdx2 mutant background, Hoxb8lacZ^+/– mice were crossed with Cdx2^+/– mice and the embryos were recovered at day 11.5 and X-gal stained. For analysis of the Hoxb8lacZ pattern in the Cdx1 single and Cdx1/Cdx2 double mutant backgrounds, Cdx1-null mice were first crossed with Hoxb8lacZ^+/– mice. Cdx1^+/–/Hoxb8lacZ^+/– mice were subsequently crossed with Cdx1-null mice or with Cdx1^+/–/Cdx2^+/– mice. The resulting Cdx1^–/–/Hoxb8lacZ^+/– and Cdx1^+/–/Cdx2^+/–/Hoxb8lacZ^+/– mice were crossed with, respectively, Cdx1^+/–/Cdx2^+/– and Cdx1-null mice, and embryos were recovered at day 8.5-12.5. The day of the plug was considered as day 0.5 of development, except when the mother had a Cdx1-null genotype. In this case, because we observed a consistent delay in development of approximately 1 day, the day after the day of the plug was considered as day 0.5.

Cdx1 genotypes were determined using PCR as described previously (Subramanian et al., 1995). Hoxb8 genotypes were determined by PCR as described elsewhere (van den Akker et al., 2001). For Cdx2 genotyping, reverse primer 5′-TAAAAGTCAACTGTGTTCGGATCC (primer 1) could be used in one PCR reaction with forward primers 5′- AGGGACTATTCAAACTACAGGAG (primer 2) and 5′-ATATTGCTGAAGAGCTTGGCGGC (primer 3). The combination of primers 1,2 and primers 1,3 makes it possible to identify the wild-type (443 bp product) and targeted (636 bp product) Cdx2 loci, respectively. Amplification conditions were denaturation at 96°C for 30 seconds, annealing at 61°C for 60 seconds and extension at 72°C for 120 seconds for 35 cycles.

Amino acid sequences of Cdx, Hox and Evx gene products were compared using Lasergene software (DNASTAR, Madison, WI).

For analysis of the Hoxb8lacZ expression patterns, day 8.5-12.5 embryos were stained with X-gal as whole mounts (Vogels et al., 1993). Paraffin wax-embedded sections of whole mount X-gal stained embryos were counterstained with Neutral Red. For analysis of the phenotypes, paraffin wax-embedded sections of day 14.5 embryos were stained with Haematoxylin and Eosin.

Newborns were stained for bone and cartilage as described previously (van den Akker et al., 2001).

Whole mount in situ hybridisation was performed as previously described (Wilkinson, 1992). The brachyury (T) probe has been described by Wilkinson et al. (Wilkinson et al., 1990). The Hoxd4 probe was from Featherstone et al. (Featherstone et al., 1988) and the Hoxb9 probe was from Graham et al. (Graham et al., 1989).

Bone and cartilage of single and compound Cdx1/Cdx2 mutant newborns (ten of each genotype) was analysed. The skeletal phenotype of Cdx1^–/– mice has been described previously (Subramanian et al., 1995). The Cdx1^–/– newborn mice that we analysed in this study showed malformations of the axial skeleton indicative of anterior transformation of vertebrae in a slightly larger AP domain than described previously. Transformations were observed from the level of the first cervical vertebra (C1; the atlas) to the level of the first lumbar vertebra (L1) (see Table 1; Fig. 1D, Fig. 2D, Fig. 3D and Fig. 4D). Heterozygous Cdx1 mutant newborns (Fig. 1C, Fig. 2C, Fig. 3C and Fig. 4C) frequently carried an eighth (pair of) rib(s) attached to the sternum (T8 towards T7 transformation, Fig. 3C) and an incomplete (pair of) rib(s) on the first lumbar vertebra (L1 towards T13 transformation, not shown, see Table 1). In addition, less frequent transformations of cervical and thoracic vertebrae were observed for this genotype (not shown, see Table 1). Skeletal abnormalities of Cdx2 heterozygous mutant mice have been described previously (Chawengsaksophak et al., 1997). The data obtained in the present study are in complete agreement with those of an earlier study, showing anterior transformation of vertebrae from the level of C6 to the level of T8 with incomplete penetrance (see Table 1, and Fig. 1B, Fig. 2B, Fig. 3B and Fig. 4B). Most frequently we observed a one- or two-sided shift in the position of the anterior tuberculum from C6 to C7, and a shift in the position of the long spinous process from T2 to T3 (Fig. 2B). Fusion of the first and second rib, and presence of an eighth pair of ribs at the sternum (Fig. 3B) were also observed in these animals but less frequently.

We found more severe skeletal defects in Cdx1^+/–/Cdx2^+/– double heterozygotes than in Cdx single heterozygous mutants (Table 1). Phenotypes that were observed frequently with incomplete penetrance in each of the single heterozygotes or only in Cdx1^+/– mice became almost fully penetrant in the double heterozygotes (Table 1 and Fig. 1B,C,E, Fig. 2B,C,E, Fig. 3B,C,E and Fig. 4B,C,E). In three of these mice, we observed anterior transformation of the first sacral vertebra (S1 to L6 transformation) and an associated shift in the position of the hindlimbs (Fig. 4E), a phenotype never observed in the single mutants. The frequency at which the transformations were observed in double heterozygotes was at all AP levels higher than the sum of the frequencies observed in each single heterozygote. At thoracic, lumbar and sacral levels, phenotypes were even observed with equal or higher frequencies in double heterozygotes than in Cdx1 homozygote mutants, suggesting equal or stronger contribution of Cdx2 at these levels. By contrast, cervical defects were found at a clearly lower frequency in the Cdx1^+/–/Cdx2^+/– mutants compared with the Cdx1^–/– mice (Table 1), indicating a predominant role of Cdx1 at rostral vertebral levels. The defects in the upper cervical area are, however, more severe in Cdx1^–/–/Cdx2^+/– double mutants than in Cdx1-null mice (Table 1 and Fig. 1F). In all cases, dorsal and lateral parts of the atlas were missing, the axis had an atlas like morphology and C3 had an axis-like morphology (Fig. 1F, Fig. 2F). In five cases, C4 also was axis-like (C4 to C2 transformation, Fig. 2F). This shows that Cdx2 does participate in patterning the upper cervical area, even if the Cdx2 mutation on its own does not lead to a phenotype at that level. In more posterior areas, a clear increase in penetrance and severity of defects was observed as well in the double mutant compared with the single and double heterozygous mutants (Fig. 2F, Fig. 3F and Fig. 4F, and see Table 1). In summary, compound Cdx1 and Cdx2 mutants exhibit anterior transformations along the complete AP axis. Cdx1^–/–/Cdx2^+/– double mutant mice exhibit the most severe alterations and their anatomical boundaries were shifted more posteriorly, as a result of these anterior transformations all along the axis.

The skeletal phenotypes of Cdx1 null mutants had been shown previously to be accompanied by posterior shifts in the anterior expression boundary of several Hox genes in paraxial mesoderm (Subramanian et al., 1995). Hoxb8 is one of the Hox genes with a rostral expression boundary at AP levels affected by the Cdx1 and Cdx2 mutations. Its expression boundaries had not been examined so far in either of the two mutants. The upper thoracic vertebrae T1 and T2, which are affected by inactivation of Hoxb8 (van den Akker et al., 1999), are also affected in Cdx1 and Cdx2 single and double mutants. In particular, fusion between the 1st and 2nd ribs was a characteristic of both Hoxb8 null animals and Cdx1/Cdx2 double mutants (Table 1). Previous work has indicated that Cdx proteins can positively regulate Hoxb8 in vivo (Charité et al., 1998). We analysed the expression of the Hoxb8lacZ ‘knock in’ allele (van den Akker et al., 1999) in Cdx1/Cdx2 single and double mutants. We examined sagittal sections of three day 11.5/12.0 embryos of each genotype. Expression boundaries were identical at day 11.5 and 12 for each genotype. Hoxb8 expression is normally maintained at low levels in prevertebra (pv) 7 and at higher levels in more posterior prevertebrae (van den Akker et al., 1999). In wild-type mice and Cdx1^+/– mice, which generally did not present any defect in the upper thorax, the expression of Hoxb8lacZ was clearly visible in pv7 (Fig. 5A,C). In Cdx1^–/– and Cdx2^+/– mutants, which show relatively mild upper thorax defects we observed either very weak or total loss of Hoxb8lacZ expression in pv7 (Fig. 5B,D). In double heterozygotes, the most anterior Hoxb8lacZ-expressing vertebra was always pv8 (Fig. 5E). In Cdx1^–/–/ Cdx2^+/– mutants, which exhibit more severe upper thorax defects, we observed either very weak expression or no expression at all in pv8, the most rostral strongly expressing vertebra being pv9 (Fig. 5F). The increasing posterior shift of the anterior boundary of Hoxb8, taken as a representative of the Hox genes with rostral boundaries in the upper thorax, correlates well with the progressively more severe abnormal upper thorax phenotypes that we observed in Cdx1Cdx2 compound mutants.

We also examined the expression domains of Hox genes with expression boundaries more anterior (Hoxd4) and more posterior (Hoxb9) than Hoxb8. Each single mutant showed a slight posterior shift in the rostral expression boundary of Hoxb9 at day 9.5 (23-29 somites), the effects being additive in the compound mutant (Fig. 6E-H). The first somite strongly expressing Hoxd4 was somite 6 for the wild type and for Cdx2 heterozygotes, and somite 7 for Cdx1 null and the compound mutants (Fig. 6A-D). The slight posterior shift of the Hoxd4 boundary in the compound Cdx mutant is therefore likely to result from the loss of the Cdx1 alleles, whereas both the Cdx1 and Cdx2 mutations contribute to the anterior restriction of the expression domains of Hoxb8 and Hoxb9.

We examined expression of the Hoxb8βgal knock in fusion protein in Cdx mutants at stages when Hoxb8 expression is still in the early establishment phase (reviewed by Deschamps et al., 1999). The Hoxb8lacZ expression level in the mesoderm at day 9 appeared to be reduced in the Cdx1^–/–/Cdx2^+/– mutants compared with Cdx1^+/– taken as a control, and the anterior boundary was detected at the level of S12 and S10/11, respectively (Fig. 7A-B). In addition, the boundary of Hoxb8lacZ expression in the neurectoderm was shifted posteriorly by the extent of one somite in a Cdx1^–/–/Cdx2^+/– double mutant embryo (and of half a somite in Cdx1^–/– and Cdx1^+/–/Cdx2^+/– double heterozygotes, not shown) compared with a stage-matched Cdx1^+/– control (Fig. 7A,B). No clear shift of the Hoxb8 expression boundary appeared to persist in the neurectoderm at later stages (day 10.5 and day 12.5, not shown). A slight shift of the Hoxb9 rostral expression boundary in the spinal cord was observed as well. This boundary was located at an AP level between somites 6 and 7 for the wild-type and Cdx2 heterozygotes, between somites 7 and 8 in Cdx1-null embryos, and in the middle of somite 8 in double mutants. The expression boundary of Hoxd4 in the neural tube was not affected by Cdx mutations in 8.5- and 9.5-day embryos (not shown).

We also investigated whether the Cdx mutations could have an effect during the initial expression of two Hox genes, Hoxb1 (at day 7.2) and Hoxb8 (at day 7.75). We could not detect any significant difference in the timing of initiation of expression of those two Hox genes in the Cdx1/Cdx2 single and compound mutant backgrounds (data not shown).

Cdx2^+/– mice have been reported to have slightly shortened tails (Chawengsaksophak et al., 1997). We counted a total number of 30-32 caudal (of 60-62 total) vertebrae in wild-type animals. Only 26-28 caudal (of 56-58 total) vertebrae were observed in Cdx2^+/– mutants. In the double heterozygotes, the number of discernible caudal vertebrae was 15-20 (of a total of 45-50), indicating a more severe truncation of the axis compared with the Cdx2^+/– heterozygotes. In Cdx1^–/–/Cdx2^+/– compound mutants, the number of discernible caudal vertebrae was further reduced to 6-11 (of a total of 36-41). In the most extreme case, only a few caudal vertebrae were observed immediately below the hindlimbs and the rest of the tail consisted of a short continuous cartilaginous structure (see Fig. 8). The severe tail truncation observed in the Cdx1^–/–/Cdx2^+/– double mutants is reminiscent of the tail defect in brachyury (T) heterozygous mice (Dobrovolskaia-Zavadskaia, 1927; Herrmann et al., 1990). Expression of T (Herrmann et al., 1990) was compared in controls and Cdx double mutants at day 7.5 and 8.5. The T expression level was not altered in Cdx double mutants. The anterior boundary of T expression, which is downregulated when somites form, was examined in several embryos of each genotype that had between 7 and 15 somites. It was always located at a distance one somite length more posteriorly than the last formed somite in both Cdx1 heterozygotes, considered as controls, and in Cdx compound mutants. The T expression domain is less extended in compound mutants compared with controls because of the loss of posteriormost tail bud tissues in the posteriorly truncated Cdx1^–/–/Cdx2^+/– mutants (Fig. 9). Because the anterior limit of T expression co-localises with the abnormal tail bending in the mutants, it seems that T expression defines the territory affected by the Cdx mutations in the tailbud. This suggests that both T and Cdx genes may be concerned with the same anatomical/morphogenetical territory but that T expression is not directly affected by the Cdx1 and Cdx2 mutations.

The peripheral nervous system is a site where phenotypical abnormalities were found in Hox loss- and gain-of-function mutants (Charité et al., 1994; van den Akker et al., 1999). We repeatedly observed abnormalities in the peripheral nervous system of Cdx mutants. Fusions between spinal ganglia were observed at cervical and thoracic levels at day 12.5 (Fig. 10A-C) and 14.5 (Fig. 10D-F). Possibly relevant to ganglia abnormalities is the fact that vertebral processes were occasionally found to be fused in Cdx mutants (Fig. 3F, Fig. 4F, Fig. 11D), suggesting the occurrence of earlier segmentation, or neural crest migration defects possibly causally connected with the fusion of spinal ganglia.

We also noticed that at least some of the Cdx1^–/–/Cdx2^+/– newborns had swollen fluid-filled abdomens (not shown). Histological analysis of the gut epithelium of double mutant and control day 13.5 embryos did not allow us to discover any meaningful alterations, and a more rigorous study of this phenotype in double mutant mice will be performed.

Examination of the appendicular skeleton revealed that one transheterozygous animal (of a total of 34), and one out of 14 Cdx1^–/–/Cdx2^+/– double mutant newborns displayed abnormal limb patterning. Digit 1 (the big toe) of one of the hindlimbs was split (Fig. 11A,B, respectively). This polydactyly is compatible with Cdx genes playing a role in patterning the lateral plate mesoderm involved in limb outgrowth, possibly in conjunction with their role in transducing AP positional information to structures along the axis via the Hox genes.

In agreement with the early expression of Cdx1 and Cdx2 in paraxial mesoderm, both genes appear to participate in patterning the vertebral column from cervical to caudal levels. While the function of Cdx2 at rostralmost levels only became manifest in the absence of Cdx1, the participation of Cdx1 at more posterior levels also became clear once one Cdx2 allele was inactivated. Combining mutations in both genes affects patterning more severely and along a more complete length of the vertebrate column than single mutations. Cdx1^–/–/Cdx2^+/– embryos exhibit the most severe phenotype, consisting in anterior homeotic-like transformations from cervical to caudal levels, implying a posterior shift of the sacrum and associated hindlimbs. In addition, the altered phenotype at caudal levels was a posterior truncation of the tail, which mildly affected Cdx2 heterozygotes but more extensively affected Cdx1^–/–/Cdx2^+/– mice. Mutations in Cdx1 and Cdx2 are accompanied by posterior shifts of Hox expression domains (Subramanian et al., 1995) (this work), and combining these mutations has a more severe effect on the Hox expression boundaries. It is therefore likely that Cdx genes affect AP patterning at least in part by regulating the Hox genes.

Cdx genes are expressed in the early CNS and dorsal root ganglia along a restricted AP domain, similar to their relatives the Hox genes. In addition, Cdx genes modulate the extent of the Hox expression domains in the CNS at early stages, as shown by the loss-of-function studies presented here, and by the gain-of-function experiments reported earlier (Charité et al., 1998). The positive regulation of Cdx gene products on Hox expression in all three germ layers was revealed by the complete loss of expression of a Hox/lacZ transgene upon mutation of the Cdx binding sites present in a crucial enhancer (Charité et al., 1998). Cdx loss-of-function mutants exhibit a slight posterior shift of the expression boundaries of at least some Hox genes in the neurectoderm. However, this effect is only observed transiently, a subsequent level of regulation probably taking over once Cdx gene expression is downregulated in this tissue. Whether the transient posterior shift of Hox gene expression in the CNS of compound Cdx mutants leads to patterning or neurological consequences is not known.

Cdx genes are expressed in the early CNS, including the neural crest precursors of the spinal ganglia. The fusion of spinal ganglia could result from a function at early stages in these neural crest progenitors, either directly or indirectly via the Hox genes. Gain- and loss-of-function Hox mutants indeed have revealed an involvement of Hox proteins in patterning spinal ganglia (Charité et al., 1994; van den Akker et al., 1999). However, Cdx newborn mutants also exhibit fusions between vertebral processes, suggesting either an earlier segmentation defect or partial sclerotome fusion. Such segmentation defects had already been reported for Cdx gain-of-function mutant embryos (Charité et al., 1998). Fusion between spinal ganglia therefore also could result from abnormal somitic properties interfering with migration of neural crest cells or causing mechanical compression of the spinal ganglia.

The expression domains of Cdx genes at early somite stages encompass the paraxial mesoderm precursors of the complete vertebral column. The early expression phase of the Cdx genes during gastrulation probably accounts for their patterning action at rostral levels of the vertebral column. The paraxial mesoderm progenitors of the cervical vertebral column (somites 5-11) are found in the epiblast lateral to the anteriormost part of the primitive streak at the late streak to head fold stages (Tam and Beddington, 1987; Tam 1988) (K. Lawson, personal communication). Cdx and Hox genes are co-expressed in the nascent paraxial mesoderm lateral to the anterior part of the streak at these embryonic stages. It is therefore likely that the Cdx/Hox regulatory interactions begin at these early, presomitic stages. Interactions between Cdx proteins and Hox genes may go on in areas of co-expression in the unsegmented paraxial mesoderm corresponding to progressively more posterior future somites/vertebrae. Concomitantly with embryonic progression, Cdx proteins would affect more posterior paraxial mesoderm, that express gradually more 5′ Hox genes. These interactions would lead to a modulation of Hox gene expression, thereby contributing to the positional identity of the somitic progenitors.

From day 8.5 onwards, these domains rapidly regress away from the sclerotomes at cervical and thoracic levels. Nevertheless, definitive vertebral patterning in the latter areas is affected by the Cdx mutations. The time of action of Cdx products instructing positional identity at rostral and trunk levels therefore must be exclusively early. In agreement with this hypothesis, Cdx mutations were shown to alter the Hox expression boundaries at early stages. It has been shown previously that an early perturbation of Hox expression boundaries could lead to altered vertebral patterning, even if the perturbation is only transient (van der Hoeven et al., 1996; Kondo and Duboule, 1999). Altogether, the data are therefore compatible with Cdx affecting early AP somitic patterning via the Hox genes. As expected from the pleiotropic effect of Cdx mutations on many Hox genes (Subramanian et al., 1995) (this work), the phenotypes in Cdx mutants extend along a more extensive AP domain than that of single Hox mutants. At a single AP level, the upper thorax, for example, the effects of mutations in Cdx1 and Cdx2 on morphogenesis were found to be co-operative, as they were on the posterior shift of the expression of a Hox gene involved in patterning at that level. According to these observations, it is possible that Cdx proteins modulate AP patterning at trunk levels exclusively via the Hox genes.

As documented in this work, Hox gene expression and AP vertebral patterning show a dose dependence on Cdx function. Instead of the five active Cdx alleles (Cdx4 is X-linked) in wild-type mice, the compound Cdx1^–/–/Cdx2^+/– mutants carry two functional Cdx alleles. A more severe impact on Hox expression and patterning can be expected in a Cdx-less situation. Nevertheless, the correlation between the loss of functional Cdx alleles and the severity of axial patterning phenotypes absolutely supports a role for Cdx genes in transduction of A-P positional information. The effects of Cdx mutations, together with the impact of Cdx experimental gain of function (Charité et al., 1998) on Hox gene expression support a role for Cdx gene products as Hox regulators. The comparatively stronger effect of Cdx deregulation on Hox gene expression in the neurectoderm in the Cdx gain-of-function transgenic experiments most probably arises from the relatively more extensive change in Cdx dose by the strong RARbeta promoter driving the transgene, than in the Cdx loss-of-function mutants. The subtlety of the effect of the Cdx mutations on the extent of the Hox expression domains may originate from the probable compensation by the remaining functional Cdx alleles. In addition, Cdx gene products are not the only regulators affecting the establishment of the Hox expression domains, and it seems that the Hox gene patterning system is well buffered against deleterious effects of mutations at one of the numerous regulatory levels.

Work on Drosophila and C. elegans has shown that cad/pal-1 is involved in posterior development and patterning only partly by regulating Hox genes. cad is the homeotic gene that mediates correct patterning of the most posterior fly segment, the analia (Moreno and Morata, 1999), which still develops normally in a HOM-less fly. pal-1 in C. elegans regulates the Hox gene mab-5, but total absence of pal-1 causes more severe, posterior-less worm phenotypes (originally called nob – no back end) than mab-5 mutations. cad therefore seems to have a homeotic function by itself. Amphioxus Cdx has been shown to belong to the ParaHox cluster, which would be historically related to the Hox cluster (Brooke et al., 1998). It is therefore likely that cad-related loci still exert a homeotic function by themselves. If this is so, mouse Cdx genes may be expected to directly pattern the most caudal embryonic structures, as cad does in the fly.

The patterning effect of Cdx genes at rostral levels is more likely to result from the regulatory action of Cdx on 3′ Hox genes than from a ‘posterior’ homeotic role of the Cdx gene products. Work in Drosophila and C. elegans has strongly suggested that Cdx gene products positively regulate several genes of the Hox cluster in the ancestral situation: cad regulates ftz in the fly (Moreno and Morata, 1999), and pal-1 regulates mab-5 and vab-7 in worms (Edgar et al., 2001). Cdx target sequences probably already existed in the ancestral Hox cluster, as witnessed by the direct transcriptional activation of mab-5 by pal-1 in the V6 cells of C. elegans (Hunter et al., 1999). In the mouse, Hox genes with rostral expression boundaries at the level of cervical to sacral levels contain potential Cdx-binding sites in their regulatory regions (Subramanian et al., 1995). The existence of this molecular crosstalk would have given Cdx gene products the possibility to regulate the 5′ Hox genes and posterior development, as well as 3′ Hox genes and more anterior patterning. Direct Cdx/Hox regulatory interactions have been observed in vertebrates (Subramanian et al., 1995; Pownall et al., 1996; Epstein et al., 1997; Isaacs et al., 1998; Charité et al., 1998). Loss of expression of a Hoxb8/lacZ transgene in mesoderm and neurectoderm upon inactivation of the Cdx-binding sites (Charité et al., 1998) may indicate a fundamental requirement of Cdx gene products in aiding trunk Hox genes to achieve their correct expression patterns. Whether the Cdx genes directly contribute positional information to paraxial mesoderm cells, or whether they transduce this information via the Hox genes is not easy to establish at this point, in the absence of total Hox disruption, or without inactivating all Cdx binding sites in the Hox clusters.

The early, maximally extending expression domain of Cdx1 corresponds to that of the most 3′ Hox genes, with a rostral expression boundary at the level of the preotic sulcus, the limit between rhombomeres 2 and 3 (Meyer and Gruss, 1993). Cdx1 and Cdx2 are initially and transiently expressed as early as Hoxb1 in the posterior part of the primitive streak at the late streak stage (Meyer and Gruss, 1993; Beck et al., 1995). These Cdx genes therefore display features of 3′-most Hox genes, in spite of the fact that they are later involved in generating and patterning posteriormost structures. According to Moreno and Morata (Moreno and Morata, 1999), cad in the ParaHox cluster might be paralogous to the 5′ neighbour of AbdB in the Hox cluster, eve. evx2 has in fact been shown to function as a posterior Hoxd gene in distal structures of the mouse limbs (Herault et al., 1996). Nevertheless, comparative analysis of the amino acid sequence of the homeodomains reveals that Cdx1 and Cdx2 are closer to Hox paralogy groups 8 and 9, and even to Hox paralogy group 1 and 2 than to the most posterior paralogy group 13 and to Evx proteins. In addition, the Cdx gene products possess a Pbx recognition motif, which is absent in 5′-most Abdb Hox proteins, such as paralogy group 13, and in Evx gene products. This motif in Cdx1 shares four of the five consensus residues with that of Hoxb4. It therefore seems that mammalian Cdx genes are relatively closely related to 3′ Hox genes, although to a lesser extent than their 3′ neighbours on the ParaHox cluster, Gsh1 and Pdx1. This could possibly explain the existence of similarities in their regulation.

Whether or not the Cdx gene products play a homeotic role on their own in the posterior part of the vertebral column, they definitely have a homeotic function along most of the axis, by modulating the position of the expression domain of Hox genes at relatively early stages.

From day 8.5/9.0 onwards, Cdx genes are not expressed any more in sclerotomes and neural tube at rostral and trunk levels, whereas these genes remain expressed in these structures at posterior levels until late embryonic stages. This late phase of Cdx expression may correspond to a different function of Cdx proteins in posterior development and patterning from the tail bud, where axial extension continues in a second phase of gastrulation (Gont et al., 1993). Persistent Cdx expression in the posteriormost part of the embryo would affect the maintenance and/or instruction of a progenitor population of tail bud-derived caudal structures (Gofflot et al., 1997; Gont et al., 1993). Strikingly, the phenotypical traits of Cdx mutants in the posterior structures are no longer homeotic-like anterior transformations but posterior truncations, much more severe in Cdx1^–/–/Cdx2^+/– than in Cdx2^+/– mutant mice. The posterior truncations in these mice are reminiscent of those found in T heterozygotes (Dobrovolskaia-Zavadskaia, 1927; Herrmann et al., 1990), Wnt5A null (Yamaguchi et al., 1999), Lef1/Tcf1 double mutants (Galceran et al., 1999), and Wnt3A (Takada et al., 1994) and Fgfr1 (Partanen et al., 1998) hypomorph mutants. This suggests that these transcription factors and signalling molecules all participate in posterior axial elongation, either by facilitating convergence/extension movements (reviewed by Sokol, 2000), or by maintaining the progenitor cell population in a proliferative state [as shown for the neural progenitors by Mathis et al. (Mathis et al., 2001)]. A function of Cdx gene products in proliferation maintenance may apply as well to the 5′ AbdB-related Hoxd and Hoxa genes, the removal of which leads to truncations of distal limb, caudal gut and external genital structures (Zákány et al., 1997; Kondo et al., 1997; Warot et al., 1997). The role of Cdx gene products in posterior axial extension, like its role in modulating AP patterning of the complete vertebral axis, cannot therefore be claimed to be either Hox-dependent or independent until the phenotype of extensive Hox deletions is available. Whatever it may be, the data presented in this paper suggest that Cdx genes influence early AP patterning all along the complete vertebral column, and act as later mediators of posterior axial elongation. It could well be that these two functions are intimately linked during the progress of morphogenesis and patterning, which are known to be interdependent, as recently shown in the analysis of the function of FgfR1 (Ciruna and Rossant, 2001).

We thank F. Meijlink and K. Lawson for critically reading the manuscript and for helpful discussions. We thank D. Duboule for the Hoxd4 probe, R. Krumlauf for the Hoxb9 probe and B. Herrmann for the T containing plasmid. The help of Carla Kroon in generating and genotyping Cdx mutant embryos was highly appreciated. We also thank one of the anonymous reviewers for suggestions concerning sequence comparisons. S. F. was supported by a Dutch ALW/NWO grant.