Msx1 is required for dorsal diencephalon patterning

In recent years, a number of experiments have demonstrated that the dorsal midline of the vertebrate neural tube is required for the specification of dorsal cell fates in the central nervous system (reviewed by Lee and Jessell, 1999). Genetic ablation of the roof plate leads to the loss of dorsal interneurons in the spinal cord (Lee et al.,2000). This results from the elimination of non-autonomous signals provided by the roof plate. Proteins from the Bmp and Wnt families are good candidates to mediate dorsal neural tube organizing activity. Thus, the double Wnt1/Wnt3a mutation results in a pronounced reduction of dorsal interneuron progenitors in the spinal cord(Ikeya et al., 1997; Muroyama et al., 2002). Similarly, Gdf7 inactivation leads to the selective loss of the D1A class of dorsal sensory interneurons (Lee et al., 1998). Furthermore, Bmp4 and Bmp7 can induce a dorsal fate in neural tube explants when added exogenously(Liem et al., 1997). Therefore, the roof plate constitutes an essential organizing center for dorsal patterning of the neural tube. Noticeably, most of these reports have dealt with the spinal cord, and little is known about the function of the dorsal midline in the more anterior part of the neural tube.

Msx genes encode homeodomain transcription factors that are expressed dorsally in the neural tube of all species throughout chordate evolution. This may therefore reflect an ancestral function for these genes in central nervous system (CNS) patterning. Indeed, the single Msx gene of Amphioxus is expressed exclusively at this site(Sharman et al., 1999). In Drosophila, Msh, the Msx ortholog, is required for the specification of lateral neuroblasts, which can be considered as the homologs of the dorsal neurons in vertebrates (Cornell and von Ohlen, 2000). In the mouse, the Msx gene family comprises three members (Msx1, Msx2 and Msx3). All three genes are expressed in the dorsal aspect of the neural tube from early stages of neurogenesis(Robert et al., 1989; Catron et al., 1996; Shimeld et al., 1996; Wang et al., 1996). At the neural fold stage, Msx1 is expressed in the future dorsal neural cells at the ectoderm/neuroectoderm boundary. Thereafter, at the neural tube closure stage, Msx1 expression is restricted, in the CNS, to the dorsal midline along its entire length(Liem et al., 1995; Furuta et al., 1997) (A.B.,Y.L. and B.R., unpublished). In the midline, Msx1 is co-expressed with Bmp and Wnt genes and might therefore play a role in signaling by these diffusible molecules. Whereas Msx3 expression is restricted to the neural tube (Shimeld et al.,1996; Wang et al.,1996), Msx1 and Msx2 are expressed in a number of structures that are formed by inductive processes between ectoderm and mesoderm, such as limb buds, craniofacial processes and tooth buds. At these sites, functional inductive interactions are required for the expression of the two genes, and these interactions involve signaling molecules of the Bmp and Fgf families (reviewed in Davidson,1995; Peters and Balling,1999).

Msx1 homozygous mutants die at birth. They exhibit cleft palate,an arrest in tooth development, with defects in the craniofacial skeleton and inner ear (Satokata and Maas,1994; Houzelstein et al.,1997). Msx2 homozygotes are viable and fertile, but display pleiotropic defects, including abnormal teeth and loss of fur that can be related to inductive processes(Satokata et al., 2000)(M.-A.N. et al., unpublished). In addition, the cerebellar lobules are reduced in number and disorganized. Analysis of Msx1/Msx2 compound mutants has further revealed functional redundancy between these genes in most of these processes because the absence of both Msx genes results in earlier and stronger phenotypes, particularly in skull bones and ectodermal organs(teeth, hair follicles and mammary glands).

The implication of Msx genes in ecto-mesodermal induction processes has been extensively studied. However, despite a number of data strongly indicative of a role for Msx genes in the development of the CNS, little is known about their role in this structure except for the implication of Msx2 in cerebellum foliation. In this report, we demonstrate, using Msx1-null mutants that we had produced previously(Houzelstein et al., 1997),that this gene is necessary for the correct development of the dorsal midline of the diencephalon. In the Msx1 mutant, the expression of genes normally expressed at the dorsal midline is extinguished specifically in prosomere 1 (P1), the most caudal subunit of the diencephalon. This leads to the downregulation of genes expressed laterally to the midline, confirming the importance of this structure for dorsal CNS patterning. Failure to form a functional dorsal midline in P1 precludes development of the subcommissural organ and leads to prenatal hydrocephalus. Wnt1 downregulation is observed early and takes place before the disappearance of the midline. This observation, together with ectopic expression experiments in the chick embryo,indicate that Wnt1 is a target of Msx1. In addition,analysis of Msx1/Msx2 double homozygous mutants confirms that the Msx genes are required for expression of Wnt1 at the pro- and mesencephalon dorsal midline. We propose a model in which Msx genes play a role in Bmp and Wnt signaling to form a functional dorsal midline. The latter acts as an organizing center to induce or maintain in neighboring tissues the expression of genes essential for dorsal brain patterning.

Generation of null alleles of the mouse Msx1 and Msx2homeobox genes by insertion of an nlacZ reporter gene and genotyping of the animals are described elsewhere(Houzelstein et al., 1997)(M.-A.N., A.B., C.R., T. Paquet, P .J. R. Barton and B.R., unpublished). Msx1 mutants were isogenized on a C57BL/6J background. Msx2mutants were on a mixed 129/Sv×C57BL/6 genetic background.

Whole-mount in situ hybridization was performed as described previously(Houzelstein et al., 1997). DNA fragments used to generate RNA probes were the following: Wnt3a,740 bp (3′ untranslated region obtained by PCR; F: GGA ATT CCA ATT TGG GCC GGA AGT CC; R: CGG GAG ATC TGA GTA TTA AGT GTC AGA GC); Otx2, 408 bp (coding for amino-acids 110 to 243); Lim1, 441 bp (coding for amino-acids 259 to 406); Pax7, Pax6 and Pax3 (complete cDNA kindly provided by Dr F. Relaix); Wnt1, 1100 bp(Parr et al., 1993) (kindly provided by Dr A. McMahon); Bmp6, 893 bp (EcoRISacI cDNA fragment kindly provided by Dr B. Hogan); and Gbx2, 620 bp (untranslated region; kindly provided by Dr G. Martin). In situ hybridization following staining for β-galactosidase activity was performed according to Tajbakhsh and Houzelstein (Tajbakhsh and Houzelstein, 1995).

For histology, embryos or newborn animals were fixed overnight in Bouin's fixative, dehydrated in ethanol, cleared in xylene and embedded in paraffin wax. Sections were cut at 15 μm and stained with Hematoxylin and Eosin.β-Galactosidase staining was performed as described by Houzelstein et al.(Houzelstein et al., 1997). Nile Blue staining was performed according to Anderson et al.(Anderson et al., 2002).

The complete coding sequence of a chick Msx1 cDNA(Robert et al., 1991) was amplified by PCR (Forward primer: CGC CGG TCG ACT GCA TGG CCC CGG CT; Reverse primer: CGC CGC TCG AGG CGG CTC GGC CCT ATG TAA). These primers were designed to add SalI and XhoI (underlined) restriction sites at the 5′ and 3′ of the Msx1 sequence, respectively. The cDNA was inserted into the XhoI site of the pCIG plasmid(Megason and McMahon, 2002)(kindly provided by Dr A. McMahon). The resulting construct was electroporated into the diencephalon of chick embryos by a modification of the procedure described by Mathis et al. (Mathis et al.,2001). Plasmid DNA was prepared using Qiagen Maxiprep and resuspended in water at 3mg/ml of DNA and 0.01% fast green. Eggs were incubated at 37°C for 40 hours to reach the 8-16 somite stage. After removal of 3 ml of albumin, a window was made into the eggshell and Hank's solution (Sigma) was dropped onto the embryo. DNA solution was introduced in the neural tube or over the head lateral ectoderm using a pressure injection apparatus (Picospritzer, General Valve). Then, four square pulses (duration=50 ms) of 25 V were applied using curved platinum electrodes (5 mm separation)placed on either side of the embryo. Embryos were recovered in PBS 6-24 hours after electroporation, photographed for GFP expression profile, then fixed for 24 hours in PFA 4%, dehydrated progressively in methanol, and conserved at-20°C. In situ hybridization was performed with a chick Wnt1probe (Bally-Cuif and Wassef,1994) as described above. Double in situ hybridization was performed using a modification of the protocol by Dietrich et al.(Dietrich et al., 1997). In brief, a chick Msx1 probe (Robert et al., 1991) was labeled with fluorescein-UTP and the chick Wnt1 probe with DIG-UTP (Roche). After hybridization, the fluorescein-labeled RNAs were detected using INT/BCIP (Roche) as substrate for the anti-fluorescein antibody-coupled alkaline phosphatase. Then, the alkaline phosphatase activity was inhibited by heating at 70°C for 30 minutes in PBS and the alkaline phosphatase-coupled anti-DIG-antibody was applied. Staining was performed with BM Purple AP substrate (Roche). The INT/BCIP red precipitate was washed out during processing of the second antibody.

We have observed that some of the Msx1 homozygous mutants display at birth a hydrocephalus, which is characterized by the dilatation of the third and fourth ventricles (A.B., Y.L. and B.R., unpublished). Anatomical analysis showed that the sub-commissural organ (SCO) is affected. The SCO is a circumventricular gland at the forebrain-midbrain boundary that originates from the neuroepithelial cells of the pretectal dorsal midline beneath the posterior commissure. Histological analysis on sagittal sections revealed that, at E15.5, the SCO was severely reduced or absent in all homozygous mutants and that the posterior commissure was disorganized(Fig. 1A,B). In addition, this commissure was colonized by numerous nucleated cells (as revealed by Hematoxylin staining) instead of being composed mainly of neural fibers(Fig. 1A,B). Morphological defects could be identified at this site from E11.5 in about one third of the homozygous mutant embryos, in the form of an indentation at the boundary between diencephalon and mesencephalon(Fig. 1C; see also Fig. 5H for E14.5).

By contrast, we did not detect any defect in the rest of the brain. In particular, the habenular recess and commissure, and the pineal gland situated anterior to the SCO region, were normal. Defects were found only between the pineal gland, which forms at the caudal boundary of prosomere 2, and the posterior commissure, which extends, caudally, to the diencephalon/mesencephalon boundary. On these anatomical criteria, all defects appear limited to structures that derive from prosomere 1 (P1)(Puelles and Rubenstein, 1993; Rubenstein and Puelles,1994).

The fate of midline Msx1-expressing cells was analyzed at earlier stages in heterozygous and homozygous mutant embryos using the nlacZreporter gene that we had targeted to the Msx1 locus(Houzelstein et al., 1997). We first verified that the β-galactosidase activity in the CNS of mutant embryos matched Msx1 transcript distribution. Similar results were obtained with X-gal staining and in situ hybridization at different stages of development, confirming that Msx1 is expressed at the midline of the CNS from, rostrally, the optic stalk to, caudally, the tip of the tail (data not shown). Analysis of sections further confirmed that, at E12.5, Msx1^nlacZ is expressed in the ependymal layer at the dorsal midline of the brain over its entire length, particularly in the SCO cells underneath the posterior commissure (data not shown).

At E14.5, all homozygous mutant embryos tested (15/15) displayed a clear interruption of the β-galactosidase domain in the midline of the pretectum, between the pineal anlage and the di-mesencephalon boundary(Fig. 2A,B). These anatomical landmarks, together with gene expression data (see below) confirm that all defects are restricted to P1. In a few cases, a short remnant of the nlacZ-expressing domain could be detected just posterior to the pineal anlage, but the main region of the midline that includes the SCO anlage was not labeled.

At E12.5, the situation was much more variable. The pretectum midline aspect varied from a total absence of β-galactosidase activity(Fig. 2C,D) to an apparently normal appearance (not shown). At E9.5, in most embryos the midline looked normal. Between E10.5 and E11.5, the phenotype appeared progressively and differed from an embryo to the other (see Fig. 6B,D). Therefore, the disappearance of the Msx1^nlacZ expression domain is of variable onset.

Analysis of the Msx1 mutant had previously not detected any difference in Msx1^nlacZ expression between the homozygous mutant and the heterozygote (Houzelstein et al., 1997). This implies that, at most sites, Msx1 is not required for its own transcription. With a more extensive analysis, we now report that the Msx1^nlacZ expression at the dorsal midline of P1 is affected in the mutant. It is unlikely that the MSX1 protein should be necessary to maintain an MSX1-positive feedback loop specifically in P1. The most plausible explanation is that the structure of the pretectum dorsal midline is affected in the absence of Msx1.

This hypothesis is further substantiated by comparing the expression profiles of several genes normally expressed in the diencephalon dorsal midline, between Msx1^-/- and wild-type or heterozygous embryos. In normal embryos, Wnt1, Wnt3a and Bmp6 are expressed in the diencephalon dorsal midline at E11.5(Parr et al., 1993; Furuta et al., 1997). In the Msx1 homozygous mutants, the expression domains of these three markers were interrupted in the pretectum midline(Fig. 3A-D and data not shown). At E12.5, Wnt3a transcripts were no longer detectable at the pretectum midline (not shown), whereas expression of several genes that are excluded from the midline was observed at this site in the Msx1mutant. This is the case for Pax6 which is normally expressed in two lateral domains adjacent to the midline(Fig. 3E; see also Fig. 5C). We further confirmed by in situ hybridization after staining for β-galactosidase in heterozygous embryos that there is no coexpression of Pax6 and Msx1^nlacZ before E14.5 (data not shown), when Pax6 begins to be expressed in the SCO layer(Estivill-Torrus et al.,2001). In the mutant, at E12.5, cells of the dorsal midline were observed to express Pax6, in such a way that the two lateral domains are no longer separated medially (Fig. 3F). At E14.5, similar results were obtained with Pax7and Lim1 as well as Pax6. Expression of these genes in the mutant, in contrast to the normal situation, extends across the midline(arrows in Fig. 5B,D,F). These results suggest that in the absence of Msx1, dorsal midline cells are not maintained in P1.

The disappearance of the dorsal midline might be due to enhanced cell death in this structure. This was evaluated using Nile Blue sulfate on E10.5-E12.5 embryos. The dorsal midline of the diencephalon is a prominent site for cell death after neural tube closure(Schlüter, 1973). In normal embryos, we indeed observed that the rate of cell death is high at the dorsal midline of the P1 prosomere relatively to the lateral as well as more rostral or caudal territories. At E10.5, Nile Blue-staining cells formed a continuous line along prosomere P1 in most embryos (8/10) and were too numerous to be counted precisely (not shown). At E11.5, dead cells were still numerous, in the range of 20-40 per embryo(Fig. 4A). Unexpectedly, at these two stages homozygous littermate embryos displayed much fewer dead cells and heterozygotes had an intermediate phenotype(Fig. 4B; summarized in Table 1). No homozygous embryo was observed with a high or intermediate dead cell rate. After Nile Blue staining, embryos were labeled for β-galactosidase activity(Fig. 4C). Only embryos that displayed an uninterrupted midline were taken into account for cell death quantitative analysis (Table 1). These showed unambiguously that decrease in cell death preceded midline disappearance, indicating that cell death was not causative in this process. At E12.5, few cells were stained with Nile Blue sulfate and no significant difference was observed between normal and mutant embryos.

In conclusion, increased cell death does not account for the disappearance of the dorsal midline in Msx1^-/- embryos. Together with the changes in gene expression at the midline (see above), these results suggest that, in the absence of Msx1, the dorsal midline changes identity to adopt more lateral territory properties, including a low rate of cell death.

We further studied the expression pattern of several genes normally expressed in dorsolateral neural tissues of P1 such as Pax6, Pax7 and Lim1 (Fujii et al.,1994; Stoykova and Gruss,1994). At E14.5, Pax6 expression is downregulated in most mutant embryos and even undetectable in some, at the midline and in the dorsolateral domain (Fig. 5C,Dand data not shown). By contrast, Pax6 expression is maintained in the epiphysis and in the telencephalon of Msx1 mutants, in accordance with our observation that defects are limited to P1 (not shown). As Msx1 and Pax6 are not co-expressed in the diencephalon of Msx1 mutants, downregulation of Pax6 must be an indirect effect of the absence of Msx1 in P1. Lim1 is less strongly expressed laterally in the rostral part of P1 in the mutant at E14.5(Fig. 5E,F). Similarly, Pax7 expression is affected in the anterior part of P1 only, with a clear rostral boundary (Fig. 5A,B). This indicates that there are two functional domains in P1 seen here for Pax7 and Lim1 expression, one that is dependent on MSX1 activity and the other not. Analysis of the expression pattern of other genes expressed in the dorsolateral part of P1, such as Otx2 and Pax3, confirms that a functional dorsal midline is necessary to specify neural tissues adjacent to the roof plate (not shown).

Strikingly, Pax7 expression profile is not affected in the mesencephalon by the Msx1 mutation(Fig. 5A,B). In situ hybridization with Gbx2, which is expressed dorsolaterally in prosomere 2, showed that the dorsolateral tissues of this prosomere are not affected either (Fig. 5G,H). Noticeably, expression which also takes place in the ventrolateral part of P1 was not affected in the Msx1^-/- mutants, indicating a lateral limit to the influence of the midline in this prosomere. These gene expression data further confirm that all defects are restricted to the dorsal part of P1 in the Msx1 mutant, and demonstrate that the loss of Msx1 function does not result in the transformation of P1 into a segment with more rostral or caudal identity.

Wnt1is downregulated before the dorsal midline disappears inMsx1^-/-embryos

Wnt1 has a dynamic pattern of expression with two major phases: at the neural plate stage, its expression takes place in a broad domain of the presumptive midbrain; from E9.5, expression is initiated at the dorsal midline of the neural tube, from the diencephalon through the spinal cord, with the exception of the r1 region in the metencephalon, and extends rostrally in the diencephalon up to the level of the P2/P3 boundary(Parr et al., 1993; Rowitch et al., 1998)(Fig. 8A). In this later phase, Msx1 and Wnt1 expression domains in the midline overlap in the spinal cord, the mesencephalon and the posterior diencephalon(Furuta et al., 1997) (A.B.,Y.L. and B.R., unpublished). As mentioned previously, in Msx1^-/- mutant embryos, at E11.5 Wnt1 transcripts were diminished in the dorsal midline cells of P1, and Msx1^nlacZ expression was also reduced. To determine whether the downregulation of Wnt1 precedes the disappearance of the Msx1^nlacZ expression domain, we performed in situ hybridization after staining for β-galactosidase activity at E10.5 and E9.5. At E10.5 in heterozygous embryos Wnt1 and Msx1^nlacZ were coexpressed throughout the mesencephalon and the posterior diencephalon (Fig. 6C). In Msx1^-/- embryos, Wnt1 and Msx1^nlacZ were co-expressed in the mesencephalon. In some Msx1^-/- embryos in which Msx1^nlacZexpression was weaker but still continuous in the P1 dorsal midline, the expression of Wnt1 was reduced and limited to a few patches of cells(Fig. 6D). At E9.5 in Msx1 mutant embryos, Msx1^nlacZ forms a continuous domain in P1 in most embryos. At this stage, Wnt1 begins to be expressed at the dorsal midline of the diencephalon(Fig. 6A). In some mutant embryos, Wnt1 expression was limited to a few scattered cells in P1(Fig. 6B). These results demonstrate that, in all cases, Wnt1 is downregulated before the Msx1^nlacZ expression domain has disappeared. This indicates that Msx1 is required for the maintenance of Wnt1expression in the midline and may also be required for its initiation. Furthermore, it raises the possibility that the disappearance of a functional midline in the dorsal domain of P1 may be secondary to the downregulation of Wnt1.

The striking overlap between Msx1 and Wnt1 expression domains in the CNS midline and the early downregulation of Wnt1 in the Msx1 mutant suggest that Msx1 may genetically interact with Wnt1. Direct interactions between the two genes have been proposed, but no firm conclusion was reached(Shang et al., 1994; Iler et al., 1995). To investigate the possibility that Msx1 might regulate Wnt1expression, we undertook electroporation of Msx1 in chick embryo diencephalon and analyzed the expression of Wnt1. A chick Msx1 cDNA (Robert et al.,1991) was inserted into the pCIG plasmid(Megason and McMahon, 2002). This plasmid contains an internal ribosome entry site followed by the coding sequence of a nuclear EGFP, downstream of the inserted sequence. This way, GFP fluorescence could be monitored after electroporation to identify cells that had incorporated plasmid DNA. Embryos were processed for in situ hybridization with a chick Wnt1 probe. Fig. 7A,B shows that ectopic expression of Msx1 in the lateral neuroepithelium of the diencephalon leads to the induction of Wnt1. Expression of Wnt1 was further efficiently induced in the telencephalon and mesencephalon (Fig. 7A-D). Wnt1 expression was always observed on the transfected side, associated with intense GFP fluorescence, and no signal was elicited by the pCIG plasmid itself (summarized in Table 2). More than 60% of treated embryos gave a positive response; the remaining ones had only a low level of GFP expression, suggesting that there is a threshold in the level of MSX1 protein that must be reached for the induction of Wnt1 (data not shown).

We further electroporated pCIG-Msx1 DNA in lateral ectodermal cells of the head. In these cells too, Msx1 induced expression of Wnt1(Fig. 7E,F). Therefore,induction of Wnt1 expression is not restricted to the neural tube. In all cases, the Wnt1 expression domain was included into the GFP-positive domain, suggesting that Msx1 induces Wnt1expression by a cell-autonomous process. To investigate this observation more rigorously, we performed in situ hybridizations successively with a chick Msx1 probe followed by a Wnt1 probe on the same electroporated embryos. A limited number of cells were induced to express Wnt1 in the Msx1 expression domain, and each of these cells was expressing a high level of Msx1(Fig. 7E,F). This confirmed that induction of Wnt1 requires a threshold level of Msx1expression. It further indicates that Wnt1 induction is very likely to be cell autonomous.

To further investigate gene interactions, we performed a kinetics of Wnt1 induction after electroporation of pCIG-Msx1 in the diencephalon(Table 2). Although GFP could be first detected 2 hours after electroporation (hae), no Wnt1induction was observed before 9 hae and it was still low at 12 hae Maximum induction was reached at 24 hae. Assuming that Wnt1 induction is cell autonomous, this leaves enough time for an indirect, intracellular process of induction, in accordance with the known properties of MSX1 as a transcriptional repressor. However, conspicuous induction requires relatively high levels of MSX1 that might not be reached during the first hours. This leaves open the possibility of a direct activation of the Wnt1 gene by MSX1.

All three Msx genes in the mouse are expressed at the dorsal midline of the neural tube. Msx1 and Msx2 are expressed in the anterior part of the CNS at early stages, while Msx3 expression does not extend rostrally beyond the isthmus(Shimeld et al., 1996; Wang et al., 1996). We reasoned that, if Msx genes are required for Wnt1 expression, the pattern of expression of the latter should be more extensively affected in the double Msx1/Msx2 mutants than in Msx1 mutants. To test this hypothesis, we took advantage of an Msx2 mutant strain constructed in our laboratory by substituting the nlacZ reporter gene to the coding sequence of the endogenous Msx2 locus, using homologous recombination(M.-A.N., C.R., T. Paquet, P. J. R. Barton and B.R., unpublished).

Compound heterozygous animals were crossed to analyze the phenotype of double homozygous mutant embryos at E10.5 and E11.5. At these stages, a number of such embryos were resorbed. The others were all smaller than normal and most had a very abnormal morphology. In some embryos, an exencephaly encompassing the posterior mesencephalon and the rhombencephalon and a spina bifida were observed (data not shown). In all cases, the brain was small and abnormal. However, in some embryos the different brain regions were clearly visible and these were analyzed for Wnt1 expression(Fig. 8B). In both E10.5 and E11.5 embryos, Wnt1 expression was totally absent from the entire diencephalon and severely reduced in the mesencephalon, where only a small caudal domain remained (Fig. 8A,B and not shown). This remnant of expression may derive from the initial expression domain of Wnt1 in the midbrain, which suggests that in the absence of functional Msx1 and Msx2, Wnt1 is not induced in the dorsal midline of the brain. In the rest of the central nervous system, where Msx3 is also expressed, Wnt1 expression looked normal.

In Msx2^nlacZ heterozygous animals, at E10.5, the Msx2^nlacZ expression domain extends over the mesencephalon and the diencephalon where it overlaps with that of Msx1^nlacZ, but the intensity of the signal was much weaker for Msx2 than for Msx1, especially in P1 (not shown). At E12.5, Msx2^nlacZ was further downregulated and almost completely undetectable in P1, while Msx1 was strongly expressed at this site (Fig. 8C,D). On the contrary, the pineal gland displayed a strong staining for the two genes(Fig. 8C,D arrowheads). The transient expression of Msx2 in P1 may explain the variation in the timing and extent of Wnt1 and Msx1^nlaczdownregulation. Our results suggest that Msx1 and Msx2 have similar functions in dorsal CNS patterning and furthermore that Msx3may be sufficient to compensate for the loss of Msx1 and Msx2 in the spinal cord and part of the rhombencephalon.

We have analyzed the effects of Msx1 loss of function in the developing central nervous system. In the mouse, all three Msx genes(Msx1, Msx2 and Msx3) are expressed in the dorsal midline of the neural tube (Robert et al.,1989; Catron et al.,1996; Shimeld et al.,1996; Wang et al.,1996). A normal roof plate fails to form in the caudal diencephalon of Msx1 mutant mice, as shown by the downregulation of genes expressed at the dorsal midline (Msx1^nlacz, Bmp6, Wnt1, Wnt3a), which can be observed from E9.5, and the apposition of expression domains normally restricted to dorsolateral regions(Pax6, Pax7, Lim1). The rate of cell death is not enhanced in the mutant midline, but instead corresponds to that observed in more lateral territories. Altogether, these results suggest that it is not midline cell survival that is affected by the mutation, but cell identity. These defects interfere with the development of structures derived from the dorsal neuroepithelium, such as the posterior commissure and the subcommissural organ, leading to hydrocephalus.

According to the neuromeric model, the diencephalon is subdivided into three anatomical subunits called the prosomeres. Prosomere 1 (P1) constitutes the most caudal subunit of the diencephalon and is located between the pineal gland and the mesencephalon. Prosomeres 2 and 3 form the dorsal and ventral thalamus, respectively. Gene expression studies have shown that prosomeres have distinct genetic identities provided by the expression of combinatory sets of genes (Puelles and Rubenstein,1993; Rubenstein et al.,1994). Furthermore, in the Pax6^sey/sey mutant,P1 is partially transformed into a mesencephalic structure(Mastick et al., 1997). This provides functional evidence in favor of the prosomeric model.

In the Msx1 mutant, abnormalities are restricted to the territory between the pineal gland and the mesencephalon, which are landmarks for the P1 territory. Interruption of the Msx1^nlacZ midline expression domain also coincides exactly with P1. Therefore, both morphological and genetic evidence indicates that the Msx1^-/- mutation affects only P1, confirming that this prosomere constitutes a genetic entity. Msx2 is expressed in the diencephalon at E10.5, but more weakly than Msx1, particularly in P1,and is further downregulated at E12.5. Therefore, Msx2 may partially compensate for the absence of Msx1 in P1 at early stages. The low residual expression of Msx2 is probably responsible for the variability in the time of onset of midline defects that we observed.

In a slightly different neuromeric model, Figdor and Stern(Figdor and Stern, 1993) have proposed that P1 is subdivided into two neuromeres: a caudal and a rostral one. In the Msx1 mutant, Pax7 and Lim1 expression in P1 is downregulated only in the more rostral region of P1. This indicates that there are, relative to the expression of these genes, two functional domains in P1, one that requires Msx1, and the other that is independent of it.

The SCO is a specialized part of the ependyma at the diencephalon dorsal midline, which develops in close association with the posterior commissure from E11. In the Msx1 mutant, the SCO is absent and the posterior commissure is affected at late stages. As Msx1 is expressed by the SCO-forming cells throughout development (P. Fernandez-Llebrez and B.R.,unpublished), further investigation is needed to determine whether Msx1 has a direct effect on the development of the SCO or whether abnormalities in the developing SCO are the consequence of more general dorsal midline defects.

Msx1 mutant mice display prenatal hydrocephalus with an incomplete penetrance (data not shown). Although the function of the SCO has not been completely characterized, prenatal hydrocephalus is correlated with failure of SCO differentiation in mouse and rat spontaneous mutants(Takeuchi et al., 1988; Perez-Figares et al., 1998). In the swaying (Wnt1^sw/sw) and small eye(Pax6^sey/sey) mutants also, the SCO fails to form and hydrocephalus is observed (Louvi and Wassef, 2000; Estivill-Torrus et al., 2001). The absence of the SCO is therefore the most likely cause of hydrocephalus in the Msx1 mutant. Furthermore, in this mutant, Wnt1 and Pax6 are downregulated and therefore crosstalk between these genes may be involved in the proper development of the SCO.

Wnt1^-/- mutant mice present a deletion of the mesencephalon and rhombomere 1 territories, leading to the apposition of prosomere 1 and rhombomere 2, and defects in ventral prosomere 2(McMahon and Bradley, 1990; Thomas and Capecchi, 1990; Mastick et al., 1996). The ventral part of P1 is not affected in the Wnt1^-/- mutant. However, the dorsal midline seems to depend on Wnt1 for its patterning, as E-cadherin, which is normally excluded from the diencephalon midline and expressed in two bands lateral to it, forms a single medial band in the Wnt1^-/- mutant(Shimamura et al., 1994). Furthermore, as previously mentioned, the subcommissural organ, a P1 midline derivative, is abnormal in the Wnt1^sw/sw mutant(Louvi and Wassef, 2000). We have shown that in the Msx1 mutant, there is a downregulation of Wnt1 and Wnt3a at E11.5. Therefore the absence of Wnt signaling may explain the disappearance of the dorsal midline.

Our data show that in Msx1 mutants, Wnt1 may be downregulated as early as E9.5, before any morphological alteration is detectable. This is the stage when the second Wnt1 expression domain,at the dorsal midline, starts to form(Parr et al., 1993; Echelard et al., 1994). Furthermore, Wnt1 downregulation just precedes the disappearance of Msx1^nlacZ-expressing cells. These observations suggest that MSX1 is required for the maintenance of Wnt1 expression in the diencephalon dorsal midline, and may be required for the activation of the second phase of Wnt1 expression. This is further supported by the analysis of the Msx1/Msx2 double mutants. At E10.5 in doubly homozygous embryos, Wnt1 expression is absent from the diencephalon and reduced in the mesencephalon. A remnant of expression caudally in the mesencephalon suggests that the early phase of Wnt1 expression has been initiated, but the second phase does not seem to take place in the absence of any functional Msx.

Msx1 has been proposed to be directly implicated in the regulation of Wnt1 expression (Echelard et al., 1994; Shang et al.,1994; Iler et al.,1995; Rowitch et al.,1998). Indeed, forced ectopic expression of Msx1 in the chick brain demonstrates that Msx1 is an inducer of Wnt1 in the neural tissue. It has been shown that a 110 bp element, which is conserved between the pufferfish Fugu rubripes and the mouse, within a 1.1 kb enhancer, is sufficient for the activation of Wnt1 expression in the mid- and forebrain (Rowitch et al.,1998). This element contains at least two homeobox-binding sites that are capable of binding MSX1 in vitro and in vivo with a high affinity,suggesting that MSX1 may interact directly with the Wnt1 promoter. However, mutation of the main binding site for MSX1 does not prevent transgene expression in the diencephalon (Iler et al., 1995). Furthermore, MSX1 has been characterized as a transcriptional repressor, and as such is unlikely to activate Wnt1directly (reviewed by Bendall and Abate-Shen, 2000). Msx1 might be involved indirectly in the regulation of Wnt1 expression, either by inducing a paracrine signaling loop or, intracellularly, by repressing a transcriptional inhibitor of Wnt1. The former hypothesis is unlikely, as expression of Wnt1 in electroporated embryos is always contained within Msx1 expression domain and strictly limited to Msx1-expressing cells. The kinetics of induction is compatible with repression of a Wnt1 repressor, as no activation of Wnt1 is observed before 9 hae. As a point of comparison, expression of Pax6,which by several criteria is a likely direct target of En2, is affected as early as 3 hae of an En2-expressing vector(Araki and Nakamura,1999).

Expression of Msx1 may be induced by several Bmps, such as Bmp2,Bmp4 and Bmp7, in different organs during embryogenesis (reviewed by Davidson, 1995). Reciprocally,Msx gene products may regulate Bmp expression. The most compelling evidence for interplay between Bmp and Msx genes is provided by the tooth germ, where Bmp4 induces Msx1 expression in the mesenchyme, which in turn is required for induction and maintenance of Bmp4 expression at this site (Chen et al., 1996). Both Bmp4 and Bmp7 have the capacity to induce roof plate cells in spinal neural plate explants, which also leads to Msx1 expression, while Bmp4 can induce ectopic Msx1 expression in lateral explants of the telencephalon (Liem et al.,1995; Furuta et al.,1997; Lee and Jessell,1999). Accordingly, application of noggin, a Bmp2-Bmp4 and Bmp7 inhibitor, to the dorsal neural plate downregulates Msx1(Muroyama et al., 2002). At the neural plate stage, Msx1 gene expression is restricted to the lateral, prospective dorsal, region of the neural folds(Robert et al., 1989) (A.B.,Y.L. and B.R., unpublished) (Liem et al.,1995), in a manner consistent with a lateral induction coming from the ectoderm which expresses Bmp4 and Bmp7. After neural tube closure, Bmp7 expression is maintained at a high level in the epidermal ectoderm of the forebrain while Bmp4 is expressed in the roof plate (Liem et al., 1995; Furuta et al., 1997). A hierarchical relationship between Bmp and Wnt genes in the dorsal neural tube has been suggested, but not definitely established. Application of Bmp4 on forebrain explants only led to a slight extension in the Wnt1expression domain (Golden et al.,1999). On the contrary, ectopic expression of a constitutively active Bmp receptor (caBmpr1a) under the control of nestin regulatory sequences led to an expansion of the Wnt1 expression domain, to include the entire alar domain of the neural tube(Panchision et al., 2001). This suggests that Bmp signaling may induce Wnt1 expression.

On the other hand, roof plate ablation has demonstrated that the dorsal midline is an essential patterning center for the dorsal neural tube(Lee et al., 2000). Roof plate cells produce signaling molecules such as Wnts or Bmps that are essential for the specification and proliferation of dorsolateral neural cells(Liem et al., 1995; Liem et al., 1997; Furuta et al., 1997; Lee and Jessell, 1999; Megason and McMahon, 2002). Mutations in the genes for these signaling factors lead to patterning defects of the dorsal CNS (Ikeya et al.,1997; Lee et al.,1998; Kim et al.,2001; Muroyama et al.,2002). In Msx1^-/- mice, several of these genes are downregulated in the dorsal midline of the diencephalon, and this is the most likely explanation for the downregulation of genes normally expressed in the dorsolateral domains.

Based on these data, a model for dorsal midline formation is proposed, in which a Bmp signal coming from the lateral/dorsal ectoderm or mesoderm induces midline expression of Msx1 at early stages(Fig. 9). This in turn is essential to maintain a functional midline that produces the diffusible signals Bmp6, Wnt1 and Wnt3a, leading to patterning of adjacent dorsolateral neural tissues. We propose that, in the case of the dorsal midline in the diencephalon, Msx genes are intermediaries between Bmp and Wnt and that this is essential for the maintenance of a functional dorsal midline.

We are very grateful to Drs P. Fernandez-Llebrez, M. Wassef, A. Pirani and A. McMahon for helpful discussions, and to Dr M. Buckingham for critical comments on the manuscript. We are indebted to Drs B. Hogan, G. Martin, A. McMahon, M. Wassef and F. Relaix for providing probes. This work was supported by the Institut Pasteur, the Centre National de la Recherche Scientifique, and grants from the Association pour la Recherche sur le Cancer and the Association Française contre les Myopathies. A.B. was the recipient of a fellowship from the Ministère de la Recherche, then from the Fondation pour la Recherche Médicale and from the Association pour la Recherche sur le Cancer. C.R. was funded by the European community, then by the Fondation pour la Recherche Médicale.