Chordin and noggin promote organizing centers of forebrain development in the mouse

The vertebrate gastrula organizer has been associated with early head development since the 1930s (Saxen,1989). The mammalian organizer, the node, promotes anterior gene expression and is required for normal development of the head primordia(Ang and Rossant, 1994;Klingensmith et al., 1999). During the initial phases of vertebrate head development, regional organizing centers generate discrete subdivisions within the anterior neural plate. One of these organizing centers, the prechordal plate (PrCP), is a mesendodermal derivative of the node (Sulik et al.,1994). The PrCP is the rostral extreme of the axial mesendoderm and underlies the rostral-most midline of the neural plate; it confers dorsoventral pattern to overlying tissue via secreted molecules such as sonic hedgehog (SHH) (Pera and Kessel,1997; Chiang et al.,1996; Shimamura and Rubenstein, 1997). Two other rostral organizers are the anterior neural ridge (ANR) and the isthmic organizer (IO). These generate pattern from within the plane of the neural ectoderm. The ANR induces and promotes forebrain character, while the IO patterns neural tissue adjacent to the midbrain/hindbrain boundary (Rubenstein and Beachy, 1998). Both of these organizing centers express diffusable factors, including FGF8(Crossley and Martin, 1995). Surgical and genetic ablation of the PrCP, ANR, IO or their effector molecules results in mispatterning and hypoplasia of the developing forebrain(Chiang et al., 1996;Houart et al., 1998;Meyers et al., 1998;Pera and Kessel, 1997;Schier et al., 1997).

BMPs also play important roles in development of the brain. Several BMPs are expressed in the mouse dorsal forebrain and facial primordia(Barlow and Francis-West, 1997;Furuta et al., 1997). Ectopic application of BMP4 to the ventral forebrain leads to holoprosencephaly in chick (Golden et al., 1999)and reduces expression of both Shh and Fgf8(Ohkubo et al., 2002). BMP4 applied to mouse forebrain explants represses anterior neural gene expression and may promote apoptosis in forebrain, hindbrain and craniofacial neural crest (Furuta et al., 1997;Graham et al., 1994;Graham et al., 1996). These observations suggest that rostral tissues require proper control of BMP activity for normal development.

The distribution and activity of BMPs in the extracellular space is regulated by the secreted factors chordin (CHRD) and Noggin (NOG), which specifically bind BMP proteins and prevent ligation of their receptors(Sasai and De Robertis, 1997). The effects of both Chrd and Nog are mediated entirely by antagonism of BMP signaling, acting upstream of signal transduction(Hammerschmidt et al., 1996;Holley et al., 1995;Holley et al., 1996). Although similar biochemical activities suggest functional redundancy, CHRD and NOG display different binding affinities for various BMPs(Piccolo et al., 1996;Zimmerman et al., 1996) and interactions with extracellular modifiers(Ashe and Levine, 1999;Piccolo et al., 1997;Scott et al., 2001). Thus,CHRD and NOG may have both shared and specific roles.

Both Chrd and Nog are expressed in the node and its derivatives (Klingensmith et al.,1999; McMahon et al.,1998). Later in embryogenesis, they are expressed in diverse structures, some uniquely and others in common(Brunet et al., 1998;Scott et al., 2000). Null mutations for mouse Chrd and Nog have been generated to assess the role of these genes in mammalian development. Chrd null homozygotes are fully viable in outbred backgrounds with partially penetrant mild defects of the chondrocranium and cervical vertebrae (D. B. and J. K. unpublished). Chrd homozygosity, however, is lethal in inbred and defined hybrid backgrounds, resulting in severe chondrocranial and pharyngeal arch defects (Bachiller et al.,2000) (R. M. A., D. B., and J. K., unpublished). Nog null homozygotes die perinatally, with defects in dorsoventral patterning and skeletal development (McMahon et al.,1998; Brunet et al.,1998). Analysis of double null homozygotes reveals functional redundancy of Chrd and Nog. Regardless of genetic background, these embryos have severe forebrain truncations in addition to defects in dorsoventral and left-right patterning(Bachiller et al., 2000) (R. M. A. and J. K., unpublished).

In this study, we have examined a second class of Chrd;Nog double mutants, Chrd^-/-;Nog^+/-, to further elucidate the functions of these genes in head development. Our results indicate thatChrd and Nog promote growth and patterning signals from the PrCP and the ANR, two organizing centers of rostral development.

Chrd^RV heterozygotes(Bachiller et al., 2000) were bred to Nog^9E heterozygotes(McMahon et al., 1998) to generate double heterozygous F₁ progeny. These were intercrossed to generate Chrd^-/-;Nog^+/- mice and were maintained on a random outbred ICR (Harlan) background. Mutant embryos were generated by timed matings of Chrd^-/-;Nog^+/-stock (Hogan et al., 1994). For comparative purposes, Chrd^+/-;Nog^+/- mice were generated and maintained in both 129Sv/J(Stottmann et al., 2001) and B6SJLF1 (Bachiller et al.,2000) genetic backgrounds. Chrd^RV andNog^9E alleles were genotyped by PCR as described previously (Bachiller et al.,2000; McMahon et al.,1998). Embryos were staged as described previously(Downs and Davies, 1993;Kaufman, 1992).

Whole-mount in situ hybridization was performed as already described(Belo et al., 1997). When necessary, staining buffer was supplemented with 5% polyvinyl alcohol (Aldrich#36, 313-8) to enhance weak signals. Standard techniques were used forβ-galactosidase staining (Hogan et al., 1994). For RT-PCR, five-somite stage embryos were dissected into three pools of tissue using glass knives: trunk, ANR and anterior midline. The ANR region excluded rostromedial tissue to avoid potential ambiguity caused by expression in the midline mesendoderm. Preparation of cDNA samples and PCR for Chrd and Hprt were as described(Stottmann et al., 2001).Nog was detected using the following primers:5′-GCATGGAGCG-CTGCCCCAGC and 5′-GAGCAGCGAGCGCAGCAGCG.

Double staining of neonatal skeletons was performed as described(McLeod, 1980). For histological sectioning, embryos were fixed overnight in Bouin's fixative(histology) or 4% paraformaldehyde (TUNEL/immunohistochemistry), and washed in PBS before embedding in paraffin wax, sectioning (8 μm) and counterstaining according to standard protocols (Hogan et al., 1994).

Nile Blue Sulfate (NBS) staining was performed as described(Trumpp et al., 1999), except embryos were incubated in Lactated Ringers containing 0.002% NBS (Sigma,N-5632) for 30 minutes at 37°C. Whole-mount TUNEL was performed as described (Conlon et al.,1995), except fragmented DNA was labeled with fluorescein-dUTP and detected with anti-fluorescein IgG (Roche). For apoptosis/proliferation assays, TUNEL was performed according to manufacturer's instructions using Fluorescein TUNEL labeling mix and TdT (Roche). Sections were then blocked in PBS containing 10% sheep serum, 1% blocking reagent (Roche) and 0.1% Tween-20 for 1 hour. Cells in metaphase were detected with anti-phosphorylated histone H3 IgG (Upstate Biotechnology) and an Alexa Red secondary antibody (Molecular Probes). Coverslips were mounted in 2.5% DABCO/90% glycerol.

At E8.5, proliferative and apoptotic indices were calculated by dividing number of metaphase-stage or TUNEL-positive cells by total number of DAPI-stained nuclei, respectively. At E9.5, indices were determined by dividing the number of labeled cells by total area of neural ectoderm. Area measurements were made using NIH image software. At least three adjacent sections were counted in each assay. Statistical significance assessed using Student's t-test.

Purified recombinant BMP2 (Genetics Institute) and BSA (Sigma) were applied to beads as described (Furuta et al.,1997). BMP2 was used at a concentration of 10 μg/ml unless otherwise indicated and BSA concentration was 1 mg/ml. Cephalic explants were isolated from five- to eight-somite embryos and contained all tissues rostral to the mid-hindbrain region. These were cultured on Nucleopore filters(Whatman) using media and culture conditions as described(Furuta et al., 1997).

Double mutants bearing null alleles of Chd and Nog were constructed to address possible functional redundancy of these BMP antagonists in mammalian development. Two types of lethal Chrd;Nog double mutant phenotypes resulted. We observed an early lethal class with severe defects in each of the three primary embryonic axes, which corresponded to the double null genotype, Chrd^-/-;Nog^-/-(Bachiller et al., 2000). We also found a perinatally lethal group of cranial-restricted mutant phenotypes,which was comprised solely of the genotypeChrd^-/-;Nog^+/-. Rarely,Chrd^-/-;Nog^+/- animals with subtle craniofacial defects were viable. Conversely, some neonates died at birth with no overt defect. However, most animals of this genotype were viable and displayed no apparent defects. Remarkably, all phenotypes are restricted to the head with no defects discernable caudal to the neck(Fig. 1). The spectrum of phenotypes reflects a syndrome of neural and craniofacial malformations collectively known as holoprosencephaly (see below). By examining the basis of these rostral defects we have investigated the roles of Chrd andNog in head development.

Cranial development in Chrd^-/-;Nog^+/-animals ranges from outwardly normal to nearly headless. Three classes of defects may be distinguished among affected mutants, defined here as midline,truncation and jaw classes. The midline class shows absence of rostroventral midline structures often resulting in cyclopia(Fig. 1B,B′, seeFig. 4P). The truncation class shows an absence of rostral-most tissues(Fig. 1C,C′). The jaw class lacks jaw elements derived from the first branchial arch and frontonasal mass (Fig. 1D-F). Although these defects are not mutually exclusive, one type usually predominates(Table 1).

In this study, we present our analysis of the initial neural defects in these mutants. Some mandibular phenotypes result from independent roles ofChrd and Nog in the first branchial arch(Stottmann et al., 2001). A study of the nature and causes of other defects in the craniofacial skeleton will be presented elsewhere (R. M. A., R. W. S. and J. K., unpublished).

Holoprosencephaly (HPE) is a deficit of ventral midline development in the forebrain. In its most severe form, a holosphere is generated: a thin-walled,singly lobed cerebral structure with a common ventricle. Less severe forms and microforms also exist with less dramatic midline deletions. HPE is frequently associated with midline craniofacial defects in humans(Ming and Muenke, 1998). Affected Chrd^-/-;Nog^+/- mice exhibit a spectrum of HPE phenotypes observed throughout gestation(Fig. 2B,E,L). Significantly,loss of medial forebrain tissue is evident before evagination of the telencephalic vesicles, the precursors to the cerebral hemispheres(Fig. 2G). In the most severe cases, HPE is evident prior to neurulation (seeFig. 4M). Truncation class mutants that lack most of the facial skeleton also entirely lack rostral brain structures. Some severely affected animals lack forebrain and midbrain(Fig. 2C,N), while more modestly affected truncation mutants lack only the forebrain(Fig. 2M). In the mildest examples, mutants have fused or absent olfactory bulbs(Fig. 2I,J). Generally, mutant neural phenotypes correlate closely with external craniofacial defects.

To reveal gestational lethality and penetrance of phenotypes inChrd;Nog mutants, we assessed embryo development during early and late gestation. Prenatal lethality is apparent only in the double homozygote(Table 2), and presence of rostral-restricted phenotypes was perfectly correlated with the genotypeChrd^-/-;Nog^+/-. Overall, HPE phenotypes were observed in 9.0% of embryos examined from E8.5-17.5 and were first detectable at the five-somite stage (∼E8.0). Furthermore,Chrd^-/-;Nog^+/- animals were under-represented by nearly 12% at weaning. This number reflects that in addition to the 9% of animals with overt HPE phenotypes, a further 3% of this mutant class die soon after birth with few, if any, detectable external defects. For example, most juvenile Chrd^-/-;Nog^+/- mice were externally normal, but some had mild midline defects such as an absence of upper incisors(n=4; data not shown).

Haploinsufficiency of Nog for forebrain development occurs only in the absence of Chrd (Table 3). However, the penetrance of these defects is low in a random-outbred genetic background. One theoretical explanation for this is that other potential genetic modifiers influence Chrd andNog function in the midline. Indeed, the mutant phenotypes observed in Chrd^-/- mice differ significantly with genetic background (see Introduction). Because of this sensitivity, we tested whetherChrd^-/-;Nog^+/- animals might have full penetrance of rostral phenotypes in defined genetic backgrounds. However, we found that most Chrd^-/-;Nog^+/- pups of both inbred (129Sv/J) and hybrid (B6SJLF1) strains exhibited no discernible holoprosencephalic phenotypes and were fully viable(Table 3, data not shown). Given the low fecundity of inbred mice and no evidence for increased phenotypic penetrance, we conducted all subsequent studies in an outbred genetic background.

To understand more fully the functions of Chrd and Nog in head formation, we characterized their expression from neural plate through early organogenesis stages (∼E7.5-8.5). The prechordal plate (PrCP) is identifiable by the co-expression of Gsc, Shh and Foxa2(Fig. 3B-D)(Belo et al., 1998;Camus et al., 2000). In situ hybridization at headfold and early somite stages reveals axial expression ofChrd and Nog through the caudal PrCP; neither is expressed in the rostral PrCP at these stages (Fig. 3A,E,I,J). Thus, the initial limit of Chrd andNog expression defines rostral and caudal subregions within the PrCP. By the five-somite stage, however, Chrd and Nog are expressed throughout the axial mesendoderm, including the PrCP(Fig. 3L,M). Chrd andNog are also clearly expressed in cephalic mesenchyme(Fig. 3L,M), and Nogis expressed near the anterior neural ridge (ANR), the boundary between surface and neural ectoderm (Fig. 3E). We did not detect Chrd in the ANR by in situ analysis; however, RT-PCR amplification demonstrates a low level ofChrd expression in the non-midline ANR, at early somite stages(Fig. 3N). In summary,Chrd and Nog are both expressed in the axial mesendoderm including the PrCP, although expression in the rostral-most PrCP is delayed until the five-somite stage. They are also expressed in or around the ANR. Thus, both genes are expressed in two of the known organizing centers of rostral head development.

Because CHRD and NOG presumably influence BMP signaling only where target BMPs are present (Piccolo et al.,1996; Zimmerman et al.,1996), we examined expression of BMP2, BMP4 and BMP7 at headfold through early somite stages (Fig. 3F-H). Each is expressed in epidermis adjacent to the neural plate. Furthermore, Bmp2 and Bmp7 are expressed in paraxial mesoderm adjacent to axial mesendoderm. Bmp7 is co-expressed withChrd and Nog in the node, notochord and caudal PrCP. Collectively, several relevant BMPs are expressed such that they may interact with CHRD and NOG, and influence PrCP and ANR activities.

Defects in Chrd^-/-;Nog^+/- pups resemble rostral holoprosencephaly phenotypes seen in Shh^-/- mice(Chiang et al., 1996). Furthermore, humans heterozygous for SHH display a range of phenotypes almost identical to that seen inChrd^-/-;Nog^+/- mice(Roessler et al., 1996). We therefore examined Shh expression and signaling inChrd^-/-;Nog^+/- mutants. In affected embryos,Shh expression is absent from the rostral mesendoderm and the rostral ventral neural midline (RVNM; Fig. 4A,B,E,F), while more caudal domains are unaffected. Patched 1(Ptch1; Ptch — Mouse Genome Informatics) andGli1 are positively regulated transcriptional targets of SHH(Lee et al., 1997;Marigo and Tabin, 1996). Therefore, expression levels reflect SHH signal transduction. Ptch1and Gli1 expression is diminished or lost in the RVNM of affectedChrd^-/-;Nog^+/- embryos(Fig. 4C,G; data not shown). SHH also induces the expression of Nkx2.1 (Titf1 —Mouse Genome Informatics) in the presumptive ventral diencephalon(Dale et al., 1997;Shimamura and Rubenstein,1997), and this expression is critical for pituitary development(Kimura et al., 1996;Takuma et al., 1998).Nkx2.1 expression is lost in affectedChrd^-/-;Nog^+/- mutants(Fig. 4D,H), resulting in aberrant infundibulum development (Fig. 4L,P). Together, these results indicate that the most rostral expression of Shh is lost in affectedChrd^-/-;Nog^+/- embryos, with a consequential diminishment of SHH signaling and PrCP function.

A total lack of SHH results in holoprosencephaly but not truncation of rostral structures (Chiang et al.,1996). We therefore suspected that an additional defect inChrd^-/-;Nog^+/- mice caused the rostral truncations. The ANR has an important role in patterning and growth of the rostral neural tube, which is mediated at least in part by FGF8 expression(Rubenstein and Beachy, 1998;Meyers et al., 1998;Shimamura and Rubenstein,1997). We therefore examined expression of Fgf8 and its downstream target Foxg1(Shimamura and Rubenstein,1997) in affected Chrd^-/-;Nog^+/-embryos. At five somites (∼E8.0), Fgf8 is expressed in bilateral domains of the ANR (Fig. 4I). However, affected Chrd^-/-;Nog^+/- embryos have reduced expression of Fgf8 in a single domain, fused at the midline(Fig. 4M). At E9.5,Fgf8 is expressed in the commissural plate and lateral epithelial domains of developing nasal placodes(Crossley and Martin, 1995)(Fig. 4J). Moderately affected embryos retain only a medial domain of rostral Fgf8 expression at dramatically reduced levels (Fig. 4N). Expression of Foxg1 is correspondingly reduced in the prosencephalon (Fig. 4K,O). These data indicate that FGF8 expression and signaling are reduced in affectedChrd^-/-;Nog^+/- embryos.

The reduction of Shh and Fgf8 expression and signaling that we have observed could be due to prior loss of cells expressing these genes. However, we have not detected morphological changes prior to the five-somite stage. Moreover, no Chrd^-/-;Nog^+/-embryos at late headfold stages showed evidence of increased apoptosis as measured by whole-mount TUNEL assay (n=41). Therefore we conclude that rostral neural ectoderm domains affected inChrd^-/-;Nog^+/- embryos are initially present at the time Shh and Fgf8 are normally first expressed. Moreover, the overall frequency ofChrd^-/-;Nog^+/- embryos displaying diminishedShh or Fgf8 was similar to the penetrance of morphological phenotypes (not shown).

Many brain and craniofacial structures are clearly absent in affectedChrd^-/-;Nog^+/- embryos by organogenesis stages. We therefore used sectioned embryos to analyze levels of apoptosis via TUNEL reaction and proliferation via anti-phosphorylated histone H3 antibody staining. At eight to 10 somites (∼E8.5), cell proliferation is greatly decreased in the forebrain of affectedChrd^-/-;Nog^+/- embryos, although apoptosis appeared comparable with wild type (Fig. 5A,B). By contrast, affected 20-25 somite (∼E9.5)Chrd^-/-;Nog^+/- embryos displayed expansion of death in the dorsal midline, lateral head mesenchyme and the trigeminal ganglion, but normal levels of neural proliferation(Fig. 5C,D). As another measure of apoptosis, we stained whole embryos with Nile Blue sulfate (NBS), which marks non-necrotic cell death. NBS staining at eight to 10 somites reveals mildly increased dorsal staining not detected in sections(Fig. 5E,F). Unexpectedly,normal ventromedial domains of apoptosis were absent, suggesting that tissues in this region have been deleted or mis-specified. At 20-25 somites, the dorsomedial stripe of NBS-positive cells was broadened in affected mutants(Fig. 5G,H). Again, the ventral midline domain of cell death was absent in affected embryos. Based on the timing of changes in gene expression relative to cell death increases elsewhere, we suspect that this loss of a normal domain of cell death is due to mis-specification of this tissue, rather than an early total deletion of the entire tissue. Collectively, these results indicate that cell proliferation is reduced in the forebrain at eight to 10 somites, following the first changes in rostral gene expression. Apoptosis levels may be mildly elevated at this stage. By 20-25 somites, proliferation levels have normalized in the forebrain, but cell death is dramatically increased in forebrain and surrounding tissues.

Apoptosis via BMP activity may be mediated by Msx transcription factors(Marazzi et al., 1997), and Msx transcription is positively regulated by BMP2 and BMP4 in the forebrain(Furuta et al., 1997). Moreover, Msx1 may repress expression of anterior neural genes(Feledy et al., 1999). We have observed increased expression of Msx1 inChrd^-/-;Nog^-/- double homozygous embryos (R. M. A. and J. K., unpublished). In addition, more severely affectedChrd^-/-;Nog^+/- embryos may display robust increases in Msx1 expression (Fig. 5I,J). However, there is no clear change in most moderately affected Chrd^-/-;Nog^+/- embryos(Fig. 5K,L), and spatial differences reflect only the absence of midline tissue. Thus, although we observe changes in apoptosis and proliferation in all affectedChrd;Nog mutants, only the more severe phenotypes also display a clear increase in Msx1 expression. Therefore it is possible thatMsx1 induction is a transient response to BMP signaling in some contexts (see more below).

Increased BMP signaling is anticipated with the loss of Chrd andNog; thus, unantagonized BMP signaling probably causes the defects seen in Chrd^-/-;Nog^+/- embryos. To test the effects of locally increased BMP activity, we cultured cephalic explants from five- to eight-somite embryos in combination with beads soaked in recombinant BMP2. Explants were prepared by placing a single bead against the ANR between the bilateral anterior neural folds (designated type 1). To confirm the presence of increased BMP signaling, we examined the expression of the direct BMP target Msx1 (Table 4, part A). Msx1 was strongly induced in regions adjacent to BMP2 beads (Fig. 6A). Strikingly, Msx1 was induced laterally at large distances (hundreds of μm) from the bead, but did not extend into the midline of the explant. This suggests that BMPs may act at long range in the ANR, and that some regions of the head are refractory to Msx1 induction by BMP signaling. We tested this by placing a BMP bead directly on the midline(designated type 2 explants). In these, the midline responded withMsx1 induction, but not in the most rostral part(Fig. 6D). Ectopic BMP2 was probably present in this region, as Msx1 induction was seen beyond this zone, in medial regions of the ANR.

Because we observed no clear increase in Msx1 expression in some affected Chrd^-/-;Nog^+/- embryos, we suspected that increased Msx1 expression may be a transient response to modestly increased BMP activity in these tissues. We used our explant assay to address the duration of Msx1 expression as a response to exogenous BMP2. Msx1 was strongly expressed in the ANR after 6 hours of culture with BMP beads (Fig. 6A). After 9 hours, expression was markedly diminished, and more so at 12 hours(Fig. 6B,C).

Because we observed a loss of Fgf8 and Shh expression in affected Chrd^-/-;Nog^+/- embryos, we tested whether increased BMP could elicit the same responses in wild-type explants(Table 4, part B). In type 1 explants, we observed suppression of Fgf8 expression in nearly all cases (93%) after 6 hours of culture using BMP2 at a concentration of 10μg/ml (Fig. 6E). Additionally, rostral domains of Shh expression were reduced in∼43% of explants. To determine whether this was due to greater sensitivity of the ANR or to bead placement, we used type 2 explants. In these, rostralShh was repressed 96% of the time. Fgf8 expression was significantly reduced but not completely repressed in 42% of explants(Fig. 6F). This demonstrates that Shh may be repressed by BMP2 in some domains without coincidentMsx1 induction. Together, these data demonstrate that effects of BMP2 on cultured cephalic explants reproduces changes in gene expression observed in affected Chrd^-/-;Nog^+/- embryos. Thus,antagonism of BMP activity is crucial both in the ANR and rostral ventral neural midline for Fgf8 and Shh expression,respectively.

If the primary cause of the defects inChrd^-/-;Nog^+/- embryos is increased BMP activity resulting from reduced BMP antagonism, explants prepared from these embryos should be more sensitive to exogenous BMP than are wild-type explants. To test this, we first compared the response of wild-type andChrd^-/-;Nog^+/- explants to BMP2-soaked beads(Table 4, part C). Here, two beads were used: one between the rostral neural folds and one against the lateral neural folds of the midbrain (type 1+). Using a BMP concentration of 10 μg/ml, Msx1 induction was consistently stronger inChrd^-/-;Nog^+/- explants than in wild type(Fig. 6G). Next, we tested whether mutant explants would show Msx1 induction at a lower BMP2 concentration than did wild-type explants. Whereas beads soaked in 10 μg/ml BMP2 strongly induced Msx1 in wild-type explants, beads prepared in 0.1 μg/ml resulted in little or no induction(Fig. 6H). However,Chrd^-/-;Nog^+/- explants responded to 0.1μg/ml BMP2-soaked beads with robust Msx1 expression(Fig. 6H). Induction ofMsx1 in Chrd^-/- explants was occasionally similar to that of Chrd^-/-;Nog^+/- explants, although a strong response in Chrd^-/- explants occurred less frequently [the relative sensitivity of different Chrd;Nog genotypes to BMPs will be presented elsewhere (R. M. A. and J. K., unpublished)]. Together, these data demonstrate thatChrd^-/-;Nog^+/- rostral tissues are sensitized to BMP signaling, strongly supporting the premise that they are deficient in BMP antagonism.

We have used compound mutants of Chrd and Nog null alleles to investigate the role of BMP antagonism in head development.Chrd^-/-;Nog^+/- mice exhibit multiple anomalies of craniofacial development, the initial defects being deficient patterning and growth of the forebrain. These problems appear to be caused by decreased activity of the PrCP and the ANR, both sites of Chrd and Nogexpression. Explant culture experiments show that exogenous BMP can elicit similar gene expression changes in wild-type explants, and thatChrd^-/-;Nog^+/- explants are greatly sensitized to BMP. These data strongly suggest that Chrd;Nog mutants suffer from too much BMP activity in rostral tissues, and that the defects observed result from this increased BMP signaling. Collectively, the results of this study reveal a novel role for BMP antagonists in promoting the functions of rostral organizing centers during forebrain development.

Molecular and genetic assays used in frog, fish and fly embryos indicate that the function of both Chordin and Noggin is to antagonize BMP signaling(Furthauer et al., 1999;Holley et al., 1995;Holley et al., 1996;Piccolo et al., 1996;Schulte-Merker et al., 1997;Zimmerman et al., 1996). Thus,genetic ablation of Chrd and Nog in mice is expected to reduce BMP antagonism. We confirmed this using two methods. First, expression of Msx1, a positive transcriptional target of BMP signaling, is upregulated and expressed ectopically in allChrd^-/-;Nog^-/- mice. This suggests increased BMP signaling in the absence of Chrd and Nog. However,ectopic Msx1 expression was observed only in the more severely affected Chrd^-/-;Nog^+/- embryos, demonstrating that one wild-type allele of Nog is largely sufficient to control BMP signaling in the rostral head. Importantly, although Chrd is transcribed at low levels after gastrulation, its expression has great functional significance. Haploinsufficiency of Nog occurs only in the absence of functional Chrd alleles.

We also assessed the response ofChrd^-/-;Nog^+/- explants to exogenous BMP. We found that cephalic explants prepared fromChrd^-/-;Nog^+/- embryos respond to BMP2 with greater induction of Msx1 than did wild-type explants. Moreover,Chrd^-/-;Nog^+/- explants exhibit robust response to BMP2 at concentrations that elicit little if any effect in wild-type explants. These results provide strong functional evidence thatChrd^-/-;Nog^+/- embryos possess reduced BMP antagonist activity. Complementary to these data, biochemical assays reveal that BMP signal transduction is increased in Chrd;Nog mutant embryos(R. M. A. and J. K., unpublished).

Midline neural deletions result from compromised function of the prechordal plate. Surgical ablation of the PrCP in mouse explants prior to four somites results in a loss of medial neural markers(Shimamura and Rubenstein,1997), as does ablation of the PrCP in chicken(Pera and Kessel, 1997). In addition, loss of signaling molecules expressed in the PrCP, such asShh in mouse (Chiang et al.,1996), results in similar neural midline defects. AffectedChrd^-/-;Nog^+/- mutants lack Shhexpression in the prechordal mesendoderm, and consequently lose expression of downstream SHH targets in surrounding tissues. Thus, the resulting midline defects in Chrd^-/-;Nog^+/- mutants are probably due to reduction of PrCP function.

At headfold stages, Chrd and Nog are expressed in the notochord and the caudal PrCP, but not in the most rostral regions; however,both genes are expressed throughout the PrCP by the five-somite stage. This early distribution of Chrd and Nog expression reveals an initial subdivision of the PrCP into two compartments. As we have observed a clear reduction of rostral-most ventral midline tissue as early as the five-somite stage, this early compartmentalization may have functional significance. In support of this hypothesis, others have demonstrated that caudal regions of rostral midline mesendoderm are required for the maintenance of identity and function of the rostral PrCP at early somite stages(Camus et al., 2000). Thus,production of CHRD and NOG protein in the caudal compartment may be essential for the activity of the rostral PrCP.

At least three BMPs bound by CHRD and NOG are expressed adjacent to the ventral midline of the rostral neural plate in mouse. In chicken, others have shown a role for BMPs in prechordal mesoderm specification(Vesque et al., 2000). Thus,the delay in Chrd and Nog expression in rostral PrCP may provide a temporal window for this to occur. Additionally, BMP7 secretion from prechordal mesoderm may be an essential co-factor with SHH in specification of rostral ventral neural midline fates; temporal regulation of Chrdexpression may limit this induction along the rostrocaudal axis(Dale et al., 1999). Although this suggests that loss of CHRD and NOG should promote rostral-ventral midline fates by increasing BMP7 signaling, the associated loss of rostral SHH may prevent induction of ectopic ventral fates in affectedChrd^-/-;Nog^+/- embryos. Alternatively, given significant differences in PrCP morphology and histology between mouse and chicken (Pera and Kessel,1997; Sulik et al.,1994), it is possible that the PrCP may function differently in mammals and birds. Taken together, our results suggest that CHRD and NOG coordinate the location, duration and composition of BMP signaling during rostral midline development.

Our study has demonstrated that application of BMP-soaked beads mimics the abnormal gene expression of Chrd^-/-;Nog^+/-mutants by repressing rostral expression of Shh in the CNS,consistent with results from chicken(Ohkubo et al., 2002). In addition to transcriptional repression of Shh, BMPs can both antagonize SHH signaling and alter the response to SHH in receiving cells(Liem et al., 1995;Murtaugh et al., 1999;Watanabe et al., 1998). Thus,in Chrd^-/-;Nog^+/- animals, ectopic BMP signaling may affect both the expression of SHH and its signal transduction.

Rostral truncation phenotypes cannot be due to impairment of SHH signaling alone, as this type of defect never occurs in Shh^-/- mice(Chiang et al., 1996) (R. M. A. and J. K., unpublished). However,Chrd^-/-;Nog^+/- mutants recapitulate rostral defects seen in some Fgf8 mutants(Meyers et al., 1998). Furthermore, Chrd^-/-;Nog^+/- mutants have reduced FGF8 expression and activity in the ANR. Taken together, these data suggest that rostral neural deletions result from impairment of ANR function. Expression of Nog in the ANR precedes Fgf8 expression, and by the five-somite stage, both Chrd and Nog are co-expressed with Fgf8 in the ANR. We have shown that several BMPs are expressed in surface ectoderm adjacent to the ANR, and that BMP2 can repressFgf8 in the ANR in vitro. Indeed, similar transcriptional repression of Fgf8 by BMPs has been observed in the chick ANR(Ohkubo et al., 2002), and in the first branchial arch of mouse(Stottmann et al., 2001). We therefore speculate that in addition to direct repression of the FGF8 targetFoxg1 by BMPs (Furuta et al.,1997), Foxg1 may be reduced indirectly by repression ofFgf8. Our results suggest that CHRD and NOG may preserve endogenous expression of FGF8 in the ANR.

Our data provide evidence for two distinct roles for Chrd andNog in patterning the rostral neural plate(Fig. 7A,B). First, CHRD and NOG secreted from the notochord and caudal PrCP antagonize the activity of BMPs expressed in the midline mesendoderm and adjacent mesodermal domains. This protects and promotes the functions of SHH and the rostral PrCP in defining the rostral neural midline. Second, CHRD and NOG secreted from the ANR antagonize BMPs that are secreted from adjacent non-neural ectoderm, and thus prevent the transcriptional repression of Fgf8. In the affected mutant embryos, reduced CHRD and NOG lead to lower BMP antagonism and thus increased BMP signaling; this in turn results in reduced SHH signaling from the rostral PrCP and FGF8 signaling from the ANR(Fig. 7C).

BMPs are reported to have proliferative, anti-proliferative and pro-apoptotic effects upon neural tissue(Furuta et al., 1997;Mabie et al., 1999;Mehler et al., 1997;Trousse et al., 2001). Our data support a role for CHRD and NOG in permitting proliferation during neurulation, and later in preventing apoptosis in the neural tube. At the eight- to 10-somite stage, cell proliferation in rostral neural ectoderm is decreased, while apoptosis is slightly increased. By 20-25 somites,proliferation in this tissue has essentially normalized, while apoptosis is substantially increased. We therefore suggest that the dramatic malformations in the forebrain of Chrd^-/-;Nog^+/-mutants are due to a combination of apoptosis-promoting and anti-proliferative activities of BMPs, together with the loss of trophic factors such as FGF8 and SHH.

The mutant phenotypes ofChrd^-/-;Nog^+/- animals showed low penetrance and wide variability. We suggest two explanations that may account for these phenomena. First, variability could result from unlinked modifier alleles that confer sensitivity to increased BMP signaling. Potential modifiers include other BMP antagonists, such as cerberus, gremlin, Dan(Nb11; Parn — Mouse Genome Informatics) and twisted gastrulation (Hsu et al.,1998; Pearce et al.,1999; Piccolo et al.,1999; Scott et al.,2001). If this were the case, the penetrance and expressivity of phenotypes in Chrd^-/-;Nog^+/- should depend strongly on genetic background. However, we observed that penetrance was low with wide phenotypic variability, even in a defined genetic background.

Alternatively, there may be natural stochastic variation in levels of BMP or BMP antagonist levels. We suggest that this variation does not approach upper or lower critical thresholds unless the balance between BMPs and antagonists is disrupted, as in Chrd;Nog mutants. Our studies withChrd^-/-;Nog^+/- cephalic explants would seem to support this idea. At a given dose of BMP2 we saw variable increases in Msx1 expression in explants of identical genotype. Consistent with this idea, mutants of several BMP signaling components have variable defects,including BMP2, BMP4, BMP7, and SMAD5 and SMAD6(Chang et al., 1999;Dudley et al., 1995;Dunn et al., 1997;Galvin et al., 2000;Winnier et al., 1995;Zhang and Bradley, 1996). This intrinsic variability in BMP signaling levels could normally be controlled by the multiple regulatory feedback loops that control BMP signaling. These include autoregulation of BMP expression levels(Ghosh-Choudhury et al., 2001;Vainio et al., 1993), as well as positive regulation by BMPs of Chrd and Nog transcription(Stottmann et al., 2001) and inhibitory SMAD expression (Afrakhte et al., 1998; Imamura et al.,1997; Nakao et al.,1997; Takase et al.,1998). Thus, regulatory mechanisms in Chrd;Nog mutants might include reduction of BMP expression and compensatory antagonism by other BMP antagonists and inhibitory SMADs.

The spectrum of phenotypes inChrd^-/-;Nog^+/- mutants mirrors the range of defects observed in human holoprosencephaly. Furthermore, human HPE is frequently associated with micrognathia, agnathia and pituitary dysgenesis,as in affected Chrd^-/-;Nog^+/- mutants. HPE is common in man, occurring in as many as one out of 250 conceptuses(reviewed by Ming and Muenke,1998). The frequency of sporadic HPE in wild-type mouse conceptuses is apparently much lower, in that we have never seen HPE in thousands of outbred embryos and fetuses examined in our laboratory (R. M. A.,A. R. L., R. W. S. and J. K., unpublished). This may imply that humans are more sensitive to environmental or genetic factors that promote HPE. Consistent with this possibility, human SHH heterozygotes can display a partially penetrant range of HPE phenotypes, while mouse Shhheterozygotes do not display HPE (Chiang et al., 1996; Ming and Muenke,1998). Overall, the variability, penetrance, and restricted location of phenotypes suggest that Chrd;Nog mutant mice reproduce many aspects of the human holoprosencephaly syndrome.

Although no BMP antagonists or BMP pathway components correspond to known HPE loci, many genes involved in human holoprosencephaly remain unidentified(Roessler and Muenke, 2001). Furthermore, it is possible that mutations in CHRD or NOGmodify one or more of the known loci. Consistent with our data, other evidence suggests that BMP signaling pathways may be involved in the pathogenesis of HPE. Addition of BMP proteins to the chick forebrain leads to reduced SHH and FGF8 expression as well as midline and anterior deletions(Golden et al., 1999;Ohkubo et al., 2002). Our genetic and embryological manipulations of mouseChrd^-/-;Nog^+/- mutants lead us to propose that insufficient BMP antagonism may underlie some human HPE by decreasing the activity of rostral organizing centers.

We thank our laboratory colleagues, M. Kirby, E. Meyers and J. Rubenstein for helpful discussions and critical reading of the manuscript, and J. Rubenstein for sharing unpublished results. S. May helped with initial characterization of mutant phenotypes. A. McMahon and R. Harland kindly provided noggin mutant mice. Probes were generously provided by R. Beddington,A. Joyner, G. Martin, P. Mitchell, J. Rossant, P. Gruss and J. Rubenstein. Recombinant BMPs were provided by Genetics Institute (Cambridge, MA). R. M. A. and R. W. S. were supported in part by a Cell and Molecular Biology NIH training grant to Duke University. This work was funded by awards to J. K. from NIDCR (R01DE13674) and the Whitehead Foundation, and to D. B. from the Victor Goodhill endowment.