Regulation of left-right asymmetry by thresholds of Pitx2c activity

Current models divide the specification of left-right asymmetry into the initial breaking of symmetry, a secondary phase that propagates an asymmetric signal along the embryo, and the last phase in which each organ interprets the cues that provide an instructive bias for asymmetric morphogenesis (Capdevila et al., 2000; Yost, 1995). It is thought that the biasing signal serves to coordinate asymmetry globally in the embryo. In the absence of bias, as in the iv mouse, asymmetric morphogenesis of each organ is uncoupled and randomized (Capdevila et al., 2000). These observations suggest that the local generation of asymmetry within each organ is an all or none, random event (Brown and Wolpert, 1990). The mechanisms that underlie the interaction of the biasing signal with each organ are poorly understood.

Insight into this problem has been obtained by the demonstration that Pitx2, a paired-related homeobox gene that was identified as the gene mutated in Rieger syndrome type I (Semina et al., 1996), plays an important role in the local generation of asymmetry within organs. Overexpression studies performed in chick and Xenopus embryos suggested that Pitx2 functioned in handed organs to interpret the asymmetric signals that originate in the pre-somitic embryo (Campione et al., 1999; Logan et al., 1998; Piedra et al., 1998; Ryan et al., 1998). Loss-of-function experiments in mice supported the idea that Pitx2 played an important role in asymmetric morphogenesis of multiple organs (Gage et al., 1999; Kitamura et al., 1999; Lin et al., 1999; Lu et al., 1999). However, it remained unclear how a single transcription factor functioned in different contexts to direct both asymmetric organ morphogenesis and the development of symmetric organs such as the teeth and eyes.

Recently, it has been recognized that the Pitx2 gene encodes three isoforms, Pitx2a and Pitx2b that are generated by alternative splicing mechanisms, and Pitx2c that uses an alternative promoter located upstream of exon 4 (Fig. 1K,L; Kitamura et al., 1999; Schweickert et al., 2000). Although overexpression of any Pitx2 isoform alters left-right asymmetric organ morphogenesis (Essner et al., 2000; Logan et al., 1998; Ryan et al., 1998), other experiments suggest that Pitx2a, Pitx2b and Pitx2c have distinct expression profiles and target genes (Essner et al., 2000; Kitamura et al., 1999; Schweickert et al., 2000). Thus, different Pitx2 isoforms may have distinct roles in left-right asymmetry and symmetric organogenesis.

To investigate the function of the Pitx2 isoforms, we used gene targeting in embryonic stem (ES) cells to generate two new Pitx2 alleles that deleted Pitx2a and Pitx2b and encoded varying levels of Pitx2c. As the developing heart and lungs express only Pitx2c, we used the new Pitx2 alleles to investigate the requirements for Pitx2c in these structures. We also investigated Pitx2c function in forming guts, which predominantly express Pitx2c. Our data reveal that asymmetric morphogenesis of the cardiac atria, lungs and duodenum have distinct requirements for Pitx2c. Moreover, our results suggest that organ-specific thresholds of Pitx2c activity play an important role in asymmetric morphogenesis.

To generate the targeting vector, we cloned the 5′ region of the Pitx2 gene using PCR amplification. Using intron-spanning oligos in exons 1 and 2, we amplified part of exon 1, the intervening intron and a small region of exon 2 that contains one putative initiator methionine (Arakawa et al., 1998; Gage and Camper, 1997; Semina et al., 1996) from a 129/Sv genomic library. The PCR product was subcloned, sequenced and restriction mapped. The 5′ end of the targeting construct, referred to as the δab^hypoc targeting vector, was generated from the PCR product, while the 3′ end was constructed from a previously characterized Pitx2 lambda phage clone (Lu et al., 1999). The δab^hypoc targeting vector, that contained approximately 6 kb of Pitx2 homologous sequences, had the IRES lacZ cassette cloned into a SalI site that was introduced by PCR into the first coding exon of Pitx2. After homologous recombination, the δab^hypoc allele resulted in deletion of the majority of exon 2 and all of exon 3. In addition, the δab^hypoc targeting vector contained the PGKneomycin resistance cassette flanked by LoxP sites that allowed removal of the PGKneomycin to generate the second Pitx2 allele, the δab allele. The δab^hypoc targeting vector was electroporated into AK7 ES cells, targeted clones identified by Southern blot, and injected into 3.5 dpc C57BL/6J mouse embryos to generate chimeras. To induce recombination between the two LoxP sites and remove the PGKneomycin cassette, we crossed δab^hypoc chimeras to the CMVCre recombinase deleter strain. The δab^hypoc and δab alleles were maintained on a mixed 129/Sv × C56BL/6J genetic background.

Whole-mount in situ hybridization was performed as previously described (Lu et al., 1999). The Pitx2c-specific probe was a 1 kb genomic fragment containing exon 4 that was linearized with XhoI and transcribed with T7 polymerase. The Pitx2a and Pitx2b-specific probe was a genomic fragment containing exons 2 and 3 that was linearized with NotI and transcribed with T3. The probes for bone morphogenetic protein 4 and sonic hedgehog have been previously described (Echelard et al., 1993; Winnier et al., 1995).

Embryos were fixed overnight in Bouin’s fixative, dehydrated through graded ethanol and embedded in paraffin. Sections were cut at 7 μm and stained with Hematoxylin and Eosin.

Whole embryo RNA was harvested with triazol reagent (Gibco) according to the manufacturers instructions. Ribonuclease protection assays were performed using the RPA II kit (Ambion). The probe for nuclease protection assays was a NcoI/NotI subclone from a Pitx2c cDNA. Pitx2c-protected fragments were quantitated with a phosphoimager and relative values analyzed for statistical significance (ANOVA). Standardization for differences in loading was performed separately using β-actin. Four experiments were performed and the difference in Pitx2c levels between δabc^null +/− and the δabc^null; δab^hypoc alleles was statistically significant (P<0.05). However, the difference in Pitx2c mRNA levels between the δabc^null +/− and δabc^null; δab and between the δabc^null; δab^hypoc and δabc^null; δab allelic combinations did not reach statistical significance.

At 8.5 dpc, Pitx2c was asymmetrically expressed in left lateral plate mesoderm and left splanchnopleure while Pitx2a and Pitx2b were expressed only symmetrically in head mesoderm (Fig. 1A,B; Kitamura et al., 1999; Schweickert et al., 2000). At 9.0 dpc, left-sided Pitx2c expression persisted in splanchnopleure and forming body wall, while at 10.5 dpc, Pitx2c was expressed in left sinus venosus, lung bud and gut (Fig. 1C,D). At these stages, Pitx2a and Pitx2b were not expressed in forming lung or cardiac structures.

In developing guts, Pitx2a, Pitx2b and Pitx2c were expressed in the stomach, cecal diverticulum, duodenum and midgut (Fig. 1D,F). With the exception of the cecal diverticulum, expression of Pitx2a and Pitx2b was a minor component of overall Pitx2 gut expression, and was most readily detectable in mice homozygous for a Pitx2 lacZ knock-in allele by X-gal staining (see below and Fig. 1F). Pitx2c, Pitx2a and Pitx2b were expressed symmetrically in oral ectoderm, body wall, umbilical structures and the developing eye at later developmental stages (Fig. 1G,H and not shown).

To dissect the functions of the Pitx2 isoforms, we used gene targeting in ES cells to generate the δab and δab^hypoc alleles that removed the Pitx2a and Pitx2b isoforms by introducing lacZ into exon 2 while deleting the coding region of exon 2 and all of exon 3 (Fig. 1K-N). The δab^hypoc allele contained a LoxP-flanked PGKneomycin cassette that was removed with Cre recombinase to create the δab allele (Fig. 1M,N).

We intercrossed δab+/− mice and found that a proportion of δab;δab mice were viable and fertile. Genotyping of weanling progeny from crosses between δab;δab and δab+/− mice showed that 29% were homozygous mutant, suggesting a loss (21%) of homozygous mutant mice in the postnatal period (n=58). Analysis of neonatal δab;δab mice revealed that postnatal lethality was secondary to cleft palate (9%; n=30) or midgut malrotation (87%; n=30) that in some cases resulted in volvulus and bowel infarction. In addition, we found that δab;δab mutants had eye defects (Fig. 1I,J).

We examined the expression pattern of Pitx2c in 10.5 dpc δab;δab mutant embryos. Pitx2c expression in left lung bud and left gut in δab;δab mutants was identical to wild-type littermate controls (Fig. 1D,E) suggesting that spatial expression of Pitx2c from the δab allele is similar to the wild-type allele. We also performed ribonuclease protection assays on whole 12.5 dpc embryo mRNA using a probe that distinguishes between Pitx2 isoforms (Fig. 1O,P) to measure Pitx2c levels in δabc^null heterozygotes and δabc^null; δab and δabc^null; δab^hypoc allelic combinations. As the δabc^null allele does not express Pitx2c (Fig. 1P, lanes 11 and 12; Fig. 1Q, lane 5), this analysis measured Pitx2c expression from the wild-type, the δab and the δab^hypoc alleles.

We found that expression of Pitx2c in the δabc^null heterozygotes was 58±11% compared with embryos with two wild-type Pitx2 alleles (Fig. 1P, lane 6; Fig. 1Q, compare lanes 1 and 2). Expression of Pitx2c in the δabc^null;δab embryos was 50±14% (not statistically significant when compared with δabc^null+/−;Fig. 1P lanes 7,8; Fig. 1Q compare lanes 2 and 3); in the δabc^null; δab^hypoc embryos, Pitx2c expression was 37±11% (P<0.05, compared with δabc^null +/−; Fig. 1P, lanes 9,10; Fig. 1Q compare lanes 2 and 4). These results show that the δab^hypoc allele encodes significantly reduced levels of Pitx2c mRNA compared with the wild-type allele in the δabc^null heterozygous embryos. Moreover, the ribonuclease protection assay data suggest a trend in which the δab allele encodes levels of Pitx2c intermediate between that of the δab^hypoc and wild-type alleles.

Although the ribonuclease protection assay analysis suggests that Pitx2c levels are comparable in the δab^hypoc and δab alleles, it is possible that this analysis has missed subtle, but biologically significant differences between the two alleles. Moreover, as it has been reported that a retained PGKneomycin cassette in an intron can interfere with Pitx2 function (Gage et al., 1999), we suspect that the PGKneomycin in the δab^hypoc locus might also have a deleterious effect on Pitx2c function.

In order to test this idea and determine if the δab^hypoc and δab alleles encode equivalent Pitx2c function, we performed a genetic experiment and intercrossed these Pitx2 alleles with the δabc^null allele. The δabc^null; δabc^null mutant embryos, that lack all Pitx2 function, died by 14.5 dpc (Gage et al., 1999; Kitamura et al., 1999; Lin et al., 1999; Lu et al., 1999). In contrast, intercrosses between the δabc^null and δab^hypoc heterozygous mice showed that about half of δabc^null; δab^hypoc embryos were still alive at 16.5 dpc (Table 1). Moreover, the δabc^null;δab mutants survived 2 days longer, as embryo loss was first detected at 18.5 dpc (Table 2). These genetic data suggest that the δabc^null; δab^hypoc and δabc^null; δab embryos survive longer than δabc^null; δabc^null embryos because of residual Pitx2c function. In addition, as a result of the retained PGKneomycin cassette, the δab^hypoc allele encodes less Pitx2c function than the δab allele. Other genetic evidence, obtained from analysis of lung phenotypes, also suggests that the δab allele encodes slightly less Pitx2c than the wild-type allele (see below).

We used the δab and δab^hypoc alleles, in conjunction with the wild-type and δabc^null alleles, to generate Pitx2 allelic combinations that encode varying levels of Pitx2c, in order to investigate the dose requirements for Pitx2c in asymmetric morphogenesis.

Although δabc^null; δabc^null mutant mice have correct rightward ventricular looping, they have numerous defects in cardiac development (Gage et al., 1999; Kitamura et al., 1999; Lin et al., 1999; Lu et al., 1999). Normal right atrial structures include the coronary sinus and the venous valves, while the left atrium has the pulmonary vein. Histological analysis showed that δabc^null; δabc^null mutants had no coronary sinus, bilateral venous valves, anomalous pulmonary venous drainage, and a deficiency in the primary interatrial septum (compare Fig. 2A,B,I with 2E-H,N,O). The outflow tract cushions of δabc^null; δabc^null mutants were symmetric and the trunks of the great arteries were malaligned and unseptated (compare Fig. 2B with 2G,H,N,P). In addition, all had defects in ventriculoarterial connections, usually double outlet right ventricle (DORV), and some had a common atrioventricular canal (Fig. 2J,P). This spectrum of morphological defects is specifically associated with right isomerism in human hearts (Brown and Anderson, 1999), and strongly suggests that the δabc^null; δabc^null embryos had right atrial isomerism.

All these cardiac defects were completely rescued in δabc^null; δab^hypoc, δabc^null; δab allelic combinations and in δab;δab mutant embryos (compare Fig. 2A,B,I,J with 2C,D,K,M). This suggests that Pitx2c is the important Pitx2 isoform for most of cardiac morphogenesis, and that only low Pitx2c levels are required. An invariable distinguishing feature between the formed left and right atrium is the morphology of the pectinate muscles (Brown and Anderson, 1999; Uemura et al., 1995). On the right hand side, the pectinate muscles are extensive and encircle the vestibule, while in the left atrium pectinate muscles are fewer and confined within the appendage. At embryonic stages, this is reflected in the degree of atrial trabeculation, with more in the right than left (Fig. 2A,I). In δabc^null; δabc^null mutant embryos, the pattern of pectinate trabeculation was reversed, being more extensive on the left than the right (Fig. 2G). As the dose of Pitx2c was increased in Pitx2 allelic combinations, pectinate trabeculation pattern was rescued. In δabc^null;δab^hypoc and δabc^null;δab embryos, there was approximately the same amount of trabeculation in right and left atria (Fig. 2L), while in δab;δab, the pattern was apparently normal (Fig. 2C). Thus, low levels of Pitx2c are sufficient for the development of most of the atrium, but higher levels are required for left identity of the atrial appendages.

Primary lung buds and mature lungs, which develop as an outpouching of the foregut, have left-right asymmetric morphology (Hogan, 1999; Fig. 3A,B). It has previously been shown that δabc^null; δabc^null embryos have complete right pulmonary isomerism (Gage et al., 1999; Kitamura et al., 1999; Lin et al., 1999; Lu et al., 1999). Consistent with the expression of Pitx2c in left splanchnopleure and primary lung buds, we found that pulmonary right isomerization was evident at the primary lung bud stage in δabc^null; δabc^null embryos (Fig. 3C,D).

We found evidence for right isomerism in all Pitx2 allelic combinations. The lungs of δabc^null; δab^hypoc and δabc^null;δab neonates showed nearly complete right pulmonary isomerism (Fig. 3E-G), suggesting that pulmonary morphogenesis is very sensitive to diminished Pitx2c function. We also found that 25% (n=16) of δabc^null heterozygous adults had partial right lung isomerization in which an extra fissure developed at the cranial aspect of the left lung (Fig. 3H). Moreover, 83% of δab; δab neonates revealed an extra fissure between what would be the cranial and middle lobes (Fig. 3I), suggesting a minor role for Pitx2a and Pitx2b in pulmonary morphogenesis. Consistent with this notion, Pitx2a and Pitx2b expression has been detected in adult lung tissue (Gage and Camper, 1997). The mild lung phenotype of the δabc^null heterozygotes, containing one copy of Pitx2c and one copy of Pitx2a and Pitx2b, compared with the severe lung phenotype of δabc^null; δab mutants, which have one copy of Pitx2c but no Pitx2a and Pitx2b, suggests that the δab allele encodes less Pitx2c function than the wild-type allele.

We investigated the secondary lung bud branching pattern in Pitx2 allelic combinations using whole-mount in situ hybridization with probes for sonic hedgehog (Shh), which marks lung bud endoderm, and bone morphogenetic protein 4 (Bmp4), which marks both endoderm and mesoderm (Bellusci et al., 1996; Bellusci et al., 1997). In δabc^null; δabc^null mutant embryos, the left-sided branching pattern was identical to that of the right lung bud or right isomerized (Fig. 3J,K,O,P). In both δabc^null; δab^hypoc and δabc^null;δab embryos, the left-sided branching pattern was also right isomerized (Fig. 3L,M,Q,R). This suggested that these Pitx2 allelic combinations fail to express adequate Pitx2c for normal left-sided branching morphogenesis. The initial aspects of branching morphogenesis in δab;δab mutant embryos were similar to wild-type, suggesting a later, minor function for Pitx2a and Pitx2b in pulmonary morphogenesis (Fig. 3N). Based on these phenotypes, and the expression pattern of Pitx2c in left splanchnic mesoderm and left primary lung bud, we conclude that high levels of Pitx2c in the lung primordia are necessary for left-specific lung morphogenesis.

The duodenum, the most rostral part of the small intestine, forms from distal foregut and rostral midgut. The early gut tube loops to the left, then as the duodenum develops, it rotates to the right to form a C-shaped structure with stereotypical relationships to the liver, pancreas and biliary tree (Moore, 1982). In all Pitx2 allelic combinations, the initial bending of the gut tube to the left was unaffected (Fig. 4A-E). However, in δabc^null; δabc^null embryos, rotation of the duodenum failed to occur in the majority (72%, n=18) of embryos examined (Fig. 4F,G). A small percentage had correct or reversed rotation at the duodenum (Table 3).

As the dosage of Pitx2c was increased in Pitx2 allelic combinations that had no Pitx2a and Pitx2b, duodenal rotation was rescued; however, orientation was initially randomized and then reversed. In δabc^null; δab^hypoc embryos, which have the lowest levels of residual Pitx2c, one half of embryos had correctly oriented duodenal rotation, while 36% had reversed rotation and 14% failed to rotate (Fig. 4H; Table 3). In contrast, the majority (83%) of δabc^null;δab embryos showed reversal of duodenal rotation (Fig. 4I; Table 3). Therefore, as Pitx2c levels gradually increased, duodenal rotation became fixed in a reverse orientation. Finally, most (88%; n=30) of δab;δab neonates and all of δabc^null heterozygotes had correct rotation of the duodenum (Fig. 4J and Table 3), showing that at these doses of Pitx2c the duodenum developed correctly.

We examined duodenal Pitx2c expression, that is normally left-sided, in δabc^null; δab and δabc^null; δab^hypoc embryos. Pitx2c was bilaterally expressed in the duodenum of these mutant embryos, suggesting the existence of a regulatory mechanism within the developing gut that normally restricts Pitx2c expression to the left side (Fig. 4K-M).

As midgut develops, it forms a cranial limb that gives rise to small bowel and a caudal limb that develops into large intestine. The midgut limbs rotate through a 270° counterclockwise movement that results in the final positioning of the small and large bowels (Moore, 1982). In 87% (n=30) of δab; δab neonates, the midgut failed to rotate resulting in a right-sided midgut mass (Fig. 4N,O). In the remainder (n=4) of δab;δab neonates midgut rotation arrested midway through rotation. In addition, 21% of δab; δab neonates showed annular pancreas (Fig. 4O).

We have used isoform-specific deletions of Pitx2a and Pitx2b that encode varying levels of Pitx2c to investigate Pitx2c function during left-right asymmetry. Our data reveal an organ-intrinsic mechanism based on differential response to Pitx2c dose for regulating asymmetric morphogenesis. Our results provide insight into the normal mechanisms regulating left-right asymmetric morphogenesis in which different developmental fields have varying requirements for Pitx2c levels.

The cardiac atria were right isomerized in δabc^null; δabc^null mutant embryos. We found right sino-atrial isomerism, including symmetry of sinus horns and bilateral paired venous valves, and abnormalities of atrioventricular and ventriculoarterial connections. In contrast, δabc^null; δab^hypoc and δabc^null;δab embryos, which encode the next highest levels of Pitx2c, had almost normal atria. From these data, we conclude that only low levels of Pitx2c are necessary to provide left identity to most of the atrium. These results also provide insight into the observation that cardiac anomalies, although described in families with Rieger syndrome, are uncommon (Bekir and Gungor, 2000; Cunningham et al., 1998; Mammi et al., 1998). In contrast, the atrial pectinate pattern requires higher levels of Pitx2c. In this respect, the atrial appendages resemble the lungs, although higher Pitx2c levels are required by the lung primordia. This fits well with the clinical observation that isomeric atrial appendages are virtually always associated with isomeric lung lobation (Brown and Anderson, 1999). In addition, the lung and atrial appendage primordia develop in close proximity.

Forming lungs require high levels of Pitx2c for normal morphogenesis. The δabc^null; δab^hypoc and δabc^null; δab allelic combinations, which had nearly normal hearts, showed strong right pulmonary isomerism phenotypes. Branching morphogenesis involves a reiterative branching mechanism in which an initial pattern is established and modified at successive steps in a stereotypical fashion (Hogan, 1999; Metzger and Krasnow, 1999). Our finding that the primary lung buds of δabc^null; δabc^null mice are right isomerized suggests that Pitx2c functions prior to or at the initial stages of the branching process. The lung phenotypes of the δab; δab mutants suggest that Pitx2a and Pitx2b has a later, more restricted function in lung morphogenesis.

The progression of duodenal phenotypes with increasing Pitx2c dosage revealed an organ-intrinsic mechanism to distinguish between randomization and reversal. The biasing model of asymmetric organ morphogenesis suggests that absence of biasing would result in organ randomization, as in the iv;iv mouse (Brown and Wolpert, 1990; Capdevila et al., 2000). Our data suggest that an intermediate level of biasing can result in reversed organ morphogenesis.

Bilateral duodenal Pitx2c expression in δabc^null; δab^hypoc and δabc^null; δab embryos suggests a mechanism within the duodenum that inhibits right-sided Pitx2c expression. This raises the possibility that relative left- versus right-sided Pitx2c levels determine the direction of gut rotation. In support of this idea, studies in Xenopus have shown that overexpression of Pitx2 on the left or right resulted in defective asymmetric morphogenesis (Essner et al., 2000). Moreover, these studies also showed that gut development was more susceptible to right-sided misexpression of Pitx2 than the heart (Essner et al., 2000), further supporting the notion of organ-specific requirements for Pitx2 function.

Our data also demonstrate that different regions of the gut are regulated by distinct mechanisms. Duodenal morphogenesis was most sensitive to changes in Pitx2c levels, while morphogenesis of the stomach was unaffected. In addition, midgut development appears to be regulated by the Pitx2a and Pitx2b isoforms, although it is possible that very high levels of Pitx2c are required for midgut development and that a slight decrease in Pitx2c expression from the δab allele is sufficient to disrupt midgut looping. Overexpression studies performed in Xenopus have also demonstrated that regional gut asymmetry can be unlinked (Bisgrove et al., 2000).

Current models propose that a biasing signal, originating at the node in mice, provides a cue that each organ uses to initiate correctly oriented asymmetric morphogenesis (Brown and Wolpert, 1990; Capdevila et al., 2000). The implication of these ideas is that asymmetric morphogenesis is all or none, either reversed or correct morphogenesis. Isomerism or loss of asymmetry would result from defects in interpretation of biasing at the organ level. Under this paradigm, uncoupling of asymmetry would be a result of defective biasing. This is illustrated by the iv and lefty1 mouse mutants that show heterotaxy (Brown and Anderson, 1999; Brown et al., 1989; Meno et al., 1998; Supp et al., 1997), implicating events occurring during the initial breaking of symmetry (iv mice) and the stabilization of left-sided gene expression by a midline barrier (lefty1 mice) in the etiology of heterotaxia. In contrast, our data suggest that left-right asymmetry within each organ is regulated not as an all-or-none decision but rather in stages. Thus, defects at the organ level, acting after biasing, can also result in heterotaxia.

We have shown that a central component of the local generation of asymmetry within an organ is a differential response to Pitx2c. These different requirements for Pitx2c dose may reflect the different morphogenetic processes, ranging from rotation of a tube to reiterated budding morphogenesis, that occur during asymmetric morphogenesis. It is conceivable that Pitx2c regulates different target genes in each organ. For example, in the atrium, Pitx2c may regulate target genes with high-affinity binding sites, while in the lung, Pitx2c target genes would have low-affinity regulatory elements. This model has the advantage of providing an understanding for how Pitx2c may regulate different morphogenetic events. Alternatively, there may exist organ-specific mechanisms to limit Pitx2c activity on a common set of target genes. This would include differential regulation of Pitx2c transcriptional levels, a mechanism that would be supported by our quantitation of Pitx2c mRNA levels. However, Pitx2c activity may also be regulated at the protein level by tight control of translation, by post-translation modification of Pitx2c or by the function of organ-specific co-factors that can modulate Pitx2c function. Further experiments are required to distinguish between these possibilities, however, the observation of a protein-protein interaction between Pitx2 and the pituitary-specific pit1 support the idea that organ-specific co-factors have a role in modulating Pitx2 function (Amendt et al., 1999).

We thank Mary Cole and Janis Smith for help with the manuscript, and R. Behringer and R. L. Johnson for comments and discussions. We thank A. Bradley, P. Soriano and R. Behringer for reagents, and A. McMahon and B. Hogan for in situ probes. Supported in part by a grant from the NIDCR (R29 DE12324) and by Basil O’Connor Starter Scholar Research Award Grant No. 5-FY97-698 from the March of Dimes to J. F. M. and the British Heart Foundation (RG/98004) to N. A. B.