FGFs and BMP4 induce both Msx1-independent and Msx1-dependent signaling pathways in early tooth development

Organogenesis is a complex process that results from a series of instructive and permissive cell-cell inductive interactions (Saxén, 1977; Wessels, 1977). In chicken embryos, morphogenetic signaling pathways have been partly elucidated in the limb bud (reviewed in Johnson and Tabin, 1997), somite (reviewed in Tajbakhsh and Sporle, 1998) and neural tube (reviewed in Jessell and Goodman, 1996). In mammalian organogenesis, however, the regulatory pathways that control inductive signaling are largely unknown.

Murine molar tooth development provides an example of an organ in which some of the early inductive interactions and molecular signaling events are beginning to be defined (Vainio et al., 1993; Chen et al., 1996; Kratochwil et al., 1996; Vaahtokari et al., 1996; Neubüser et al., 1997; Jernvall et al., 1998; reviewed in Thesleff et al., 1995; Maas and Bei, 1997; Thesleff and Sharpe, 1997). In the mouse embryo, molar tooth development commences morphologically at mouse embryonic day 11.5 (E11.5), with a thickening of the dental epithelium to form the dental lamina. Before E12.5, the dental epithelium can elicit tooth formation when recombined with neural-crest-derived second branchial arch mesenchyme, while the reciprocal combination fails (Mina and Kollar, 1987; Lumsden, 1988). Hence, at this time, the tooth-forming inductive potential resides in the prospective dental epithelium. The epithelium invaginates to form a tooth bud in the underlying dental mesenchyme, which proliferates and condenses. Subsequent to E12.5, dental mesenchyme can induce tooth formation when recombined with second arch epithelium, while recombinants containing dental epithelium and second arch mesenchyme fail. Thus a shift in odontogenic potential from dental epithelium to dental mesenchyme occurs at E12.5.

Recent studies have suggested that the TGF-β superfamily member Bone Morphogenetic Protein 4 (BMP4) constitutes one component of the inductive signal that transfers tooth inductive potential from dental epithelium to mesenchyme (Vainio et al., 1993). Bmp4 expression is first observed in the molar tooth at E11.5 in the dental lamina epithelium but then shifts at E12.5 to the dental mesenchyme, coincident with the shift in tooth developmental potential between tissue layers and BMP4 can induce morphologic changes in dental mesenchyme (Vainio et al., 1993; Turecková et al., 1995). Msx genes are also implicated in the epithelial-mesenchymal interactions involved in tooth development. Msx1 is strongly expressed in the dental mesenchyme throughout the lamina, bud, cap and bell stages of odontogenesis (MacKenzie et al., 1991a,b, 1992; Maas et al., 1996). Msx2 expression is initially restricted to the mesenchyme directly beneath the prospective dental lamina, thereafter localizing to the dental papilla mesenchyme and the epithelial enamel knot (MacKenzie et al., 1992). The involvement of Msx gene function in tooth development is demonstrated by Msx1 knockout mice, which exhibit a highly penetrant arrest at the bud stage of molar tooth development (Satokata and Maas, 1994).

Insight into the genetic relationship between Msx1 and Bmp4 comes from experiments showing that Bmp4 expression is reduced in the Msx1 mutant tooth mesenchyme but is preserved in Msx1 mutant epithelium (Bei et al., 1996; Chen et al., 1996). These results indicate that Msx1 is required for the expression of Bmp4 in the dental mesenchyme and that Bmp4 therefore functions downstream of Msx1 in the dental mesenchyme. On the contrary, epithelial Bmp4 expression does not require Msx1 for its expression and therefore acts upstream of Msx1. Experiments have shown that BMP4 can induce the expression in the dental mesenchyme of Msx1 and its own expression (Vainio et al., 1993). However, in Msx1 mutant dental mesenchyme, BMP4 cannot induce its own expression indicating that mesenchymal Bmp4 expression requires Msx1 function. Furthermore, addition of recombinant BMP4 to chemically defined media partly rescues the Msx1 mutant tooth bud phenotype to the cap stage of odontogenesis (Chen et al., 1996), further substantiating the view that mesenchymal Bmp4 functions downstream of Msx1 and suggesting that mesenchymal BMP4 acts back upon the dental epithelium to mediate the reciprocal epithelial-mesenchymal interactions that occur during tooth morphogenesis.

Nonetheless, despite the fact that epithelial BMP4 is able to induce its own expression and that of Msx1 in the dental mesenchyme, BMP4 cannot substitute for all the inductive functions of the dental epithelium. For example, dental epithelium, but not recombinant BMP4, induces the expression of the heparan sulfate proteoglycan syndecan-1 and cell proliferation of dental mesenchyme (Jernvall et al., 1994). This suggests that other epithelial factors besides BMP4 are responsible for the induction of cell proliferation and of syndecan-1 gene expression. Plausible candidates for this role are members of the fibroblast growth factor family (FGFs). For example, FGF4 can stimulate cell proliferation of dental mesenchyme (Jernvall et al., 1994) and FGF4 and other FGFs can induce syndecan-1 in the dental mesenchyme (Chen et al., 1996). However, Fgf4 is expressed in the dental epithelium at E14.5 in the enamel knot, too late to be the natural inducer of dental mesenchyme at the lamina stage. In contrast, Fgf1, Fgf2, Fgf8 and Fgf9 are expressed in the early molar dental lamina, and FGF8 can induce a translucent zone in dental mesenchyme and has been proposed to act antagonistically with BMP4 to specify the sites of tooth initiation (Cam et al., 1992; Heikinheimo et al., 1994; Neubüser et al., 1997; Kettunen and Thesleff, 1998). In the dental mesenchyme, another member of the FGF family, Fgf3, has also been found to be expressed as early as E12.5 (Thesleff and Vaahtokari, 1992). However, the relationship of epithelial FGF signaling and mesenchymal Fgf3 expression to Msx function during early tooth development remains unknown.

Targeted mutations in either of two members of the distal-less homeobox gene family, Dlx1 or Dlx2, affect development of skeletal elements derived from the proximal ends of first and second branchial arches (Qiu et al., 1995, 1997) but have no effect on tooth development. However, mice compounded for mutations in both Dlx1 and Dlx2 exhibit a selective absence of upper molars and an arrest at the lamina stage (Thomas et al., 1997). This has been explained on the basis of functional redundancy with other Dlx genes expressed in mandibular but not maxillary mesenchyme. Although mesenchymal Bmp4 and Msx1 expression are preserved in the arrested maxillary molar tooth germs in Dlx1,Dlx2 knockout mice (Thomas et al., 1997), less is known about how Dlx genes might fit into the genetic pathway that includes Msx1 and Bmp4.

Here we show that FGFs that are expressed in the dental epithelium can induce Fgf3 expression in the dental mesenchyme via a Msx1-dependent mechanism distinct from that by which epithelial BMP4 induces its own gene expression in dental mesenchyme. The data indicate that Msx1 is necessary but not sufficient for the expression of inductive signaling molecules in the dental mesenchyme. In addition, the activation of Dlx1 and Dlx2 expression is preserved in Msx1 mutant tooth germs, and Msx1,Msx2 double mutants exhibit a tooth phenotype that resembles that in Dlx1,Dlx2 double mutants in terms of stage of developmental arrest. FGF8 and BMP4 differentially induce Dlx1 and Dlx2 expression in the dental mesenchyme in a Msx1-independent fashion, further supporting the conclusion that Dlx and Msx genes act in parallel during early induction of the dental mesenchyme in response to epithelial signals.

Embryos were collected from matings of Msx1^+/− × Msx1^+/− mice or Msx1,Msx2 double heterozygous mice maintained in a N5-6 BALB/c background, and fixed in 4% paraformaldehyde/phosphate-buffered saline (PBS) for approximately 5 hours, dehydrated through increasing concentrations of ethanol and embedded in paraffin wax. The preparation, genotyping and detailed phenotype analysis of the Msx2 mutant mice and Msx compound mutant embryos will be described in detail elsewhere (R. Maas, unpublished data). Genotyping of Msx1 mutant mice was performed as previously described (Chen et al., 1996).

Murine Fgf3 and Fgf8 probes (gifts from Gail Martin, University of California at San Francisco, California), and Dlx1 and Dlx2 probes (gift from John L. R. Rubenstein, University of California, San Francisco, CA) were used. A 1300 bp fragment of murine Fgf3 cDNA subcloned into pBluescript was digested with BssHII and transcribed with T7 RNA polymerase for an antisense probe. An 800 bp fragment of murine cDNA for Fgf8 subcloned into pBluescript SK+ was digested with PstI and transcribed with T7 RNA polymerase for an antisense probe. A 240 bp fragment of murine cDNA for Dlx1 subcloned into pBS KDD was digested with BamHI or EcoRV and transcribed with T7 or T3 RNA polymerase for antisense or sense probes, respectively. A 560 bp fragment of murine cDNA for Dlx2 subcloned into a variant of pBS SK− (E61) plasmid was digested EcoRI and NotI and transcribed with T3 and T7 RNA polymerase for antisense and sense probes, respectively. The RNA probes were radiolabeled with [α-³⁵S]UTP or digoxigenin using T3 or T7 RNA polymerase. The probes were reduced to an average size of 100-150 bp by hydrolysis.

Bead implantation and tissue recombination experiments were performed according to previously described procedures (Vainio et al., 1993; Chen et al., 1996). For bead implantation, Affi-Gel blue agarose beads (100-200 mesh, 75-150 μm diameter, Bio-Rad) were incubated with 70-100 ng/μl recombinant human BMP4 protein (Genetics Institute, Cambridge, MA) at 37°C for 30 minutes, or heparin acrylic beads (Sigma, St Louis, MO) were incubated recombinant human FGF1 (700 ng/μl), human FGF2 (100 ng/μl) or mouse FGF8 (250 ng/μl) (R&D Systems, Minneapolis, MN) proteins at 37°C for 1 hour. Control beads were soaked with similar concentrations of BSA under the same conditions. Protein-soaked beads were stored at 4°C and used within 1 week. Freshly isolated dental mesenchymes were placed on Nuclepore filters (pore size, 0.1 mm), and protein-soaked beads were washed in PBS and placed on the top of the mesenchyme. All explants were cultured on the filters, supported by metal grids in Dulbecco’s minimal essential medium with 10% FCS at 37°C for 24 hours. After culture, explants were fixed and processed for whole-mount in situ hybridization.

Whole-mount in situ hybridization experiments were performed as previously described (Xu et al., 1997). For tissue section in situ hybridizations with α-³⁵S-labeled probes, 7 μm sections were cut from wax-embedded embryos, dewaxed through xylene, rehydrated and refixed in 4% paraformaldehyde/phosphate-buffered saline. Further section preparation and hybridization were then performed as described (Sassoon and Rosenthal, 1993). Slides were dipped in Kodak NTB2 radiographic emulsion diluted 1:1 with distilled H₂O. Following 6-10 days of exposure at 4°C the slides were developed using Kodak D19 developer and fixed using Kodak fixer and counterstained in Hoëscht fluorescent dye. Photographs were taken using Kodak EPY 64T film on Zeiss Axiophot microscope using dark-ground illumination.

Histology was performed by fixation in 10% formalin or 4% paraformaldehyde followed by dehydration and embedding in paraffin. Sections were cut by microtome at 7 μm and stained with hematoxylin and eosin.

To investigate the relationship between Msx1 function and Fgf expression in early tooth development, Fgf8 and Fgf3 expression in wild-type and Msx1-deficient molar tooth germs was examined by in situ hybridization (Fig. 1). Fgf8 expression is maintained in E11.5 dental epithelium in Msx1 mutant embryos, indicating that epithelial expression of Fgf8 does not require Msx1 (Fig. 1A,B). In contrast, at E13.5, corresponding to the stage at which tooth development arrests in the Msx1 mutants, Fgf3 expression was undetectable in the Msx1-deficient molar tooth mesenchyme (Fig. 1C,D). Some Msx1 mutant tooth buds arrest earlier than E13.5; however, in situ hybridization experiments performed at E12.5 confirmed the above result (data not shown). Thus, while Fgf8 expression in the dental epithelium does not require Msx1, Fgf3 expression in the dental mesenchyme is Msx1 dependent.

To determine if FGFs can induce their own expression in dental mesenchyme and, if so, whether this induction requires Msx1, microdissected wild-type and Msx1 mutant E11.5 molar mesenchymes were implanted with beads soaked in recombinant FGF1, FGF2 or FGF8. Following organ culture for 24 hours, the specimens were analyzed for Fgf3 expression by whole-mount in situ hybridization (Fig. 2; Table 1). Experiments employing concentrations between 100 and 700 ng/μl for all three FGF-soaked beads gave similar results. FGF1, FGF2 and FGF8 all induce Fgf3 expression in wild-type dental mesenchyme (Fig. 2A,C; Table 1). Experiments combining wild-type dental epithelium with wild-type dental mesenchyme also showed induction of Fgf3 expression in the dental mesenchyme (Fig. 2A, arrow). However, while FGF1, FGF2 and FGF8 and dental epithelium all induced Fgf3 expression in wild-type dental mesenchyme, none of these FGFs induced Fgf3 expression above background in Msx1 mutant dental mesenchyme (Fig. 2B,D; Table 1). Beads soaked in recombinant FGF8 were also tested, as a positive control, for their ability to induce Msx1 expression in wild-type dental mesenchyme. As described (Kettunen and Thesleff, 1998), FGF8 induces Msx1 expression in wild-type dental mesenchyme (data not shown). Conversely, control experiments using beads containing BSA or sense riboprobes gave minimal if any signal (arrowhead in Fig. 2C and data not shown, respectively). These results indicate that Msx1 is required for FGFs to induce Fgf3 expression in dental mesenchyme.

Beads soaked in recombinant FGFs are not able to induce Bmp4 expression in wild-type dental mesenchyme (data not shown, and Chen et al., 1996). To test whether BMP4 can induce Fgf3 expression in dental mesenchyme, BMP4 beads were implanted in wild-type dental mesenchyme and the explants assayed by whole-mount in situ for Fgf3 expression. Similar to control experiments using BSA beads, BMP4 beads cannot induce Fgf3 expression in wild-type dental mesenchyme (Fig. 2E,F; Table 1). Thus, the BMP4- and FGF-signaling pathways are independent with respect to the induction of Bmp4 and Fgf3 expression in the dental mesenchyme. Since BMP4 and FGF8 can each induce Msx1 expression and Msx1 is necessary for Fgf3 and Bmp4 expression in dental mesenchyme, the failure of BMP4 and FGFs to respectively induce Fgf3 and Bmp4 expression indicates that Msx1 is necessary but not sufficient for expression of these downstream genes in dental mesenchyme. Therefore, other factors may interact with the Msx1 gene product, directly or indirectly, to provide sufficiency for downstream gene expression.

Members of the distal-less gene family could act coordinately with Msx in tooth induction, since Msx and Dlx homeoproteins interact in vitro (Zhang et al., 1997) and Dlx1,Dlx2 double mutants exhibit a lamina stage arrest in maxillary molar tooth development which, although earlier and more restricted than the bud stage arrest in Msx1 mutants, could nonetheless be developmentally related (Thomas et al., 1997). To test whether Msx1 is required for Dlx1 and Dlx2 expression in the dental mesenchyme, in situ hybridization was performed in wild-type and Msx1 mutant upper and lower molar mesenchymes (Fig. 3). At E11.5 Dlx1 expression was present in Msx1 mutant dental mesenchyme, indicating that early mesenchymal Dlx1 expression does not require Msx1 (Fig. 3A,B). At E11.0, Dlx2 expression in both dental epithelium and mesenchyme was observed in Msx1 mutant embryos (data not shown) and, at E11.5, Dlx2 expression continued to be present in Msx1 mutant dental mesenchyme at wild-type levels (Fig. 3C,D). At the bud stage of tooth development, E12.5-13.5, Dlx1 expression was maintained at control levels in the Msx1 mutant dental mesenchyme (Fig. 3E,F, shown for E13.0). In contrast, in both Msx1 mutant upper and lower molar tooth germs, Dlx2 expression was not detected (Fig. 3G,H; arrows). Thus, Msx1 is dispensable for the initial expression of Dlx1 or Dlx2 in dental mesenchyme at the lamina stage of tooth development while, at the subsequent bud stage, Dlx2 mesenchymal expression but not that of Dlx1 requires Msx1.

To further test whether Msx and Dlx genes might act in parallel and whether functional redundancy between Msx1 and Msx2 might mask an earlier requirement for Msx function during tooth initiation, tooth development in Msx1,Msx2 compound mutant embryos was analyzed histologically (Fig. 4). At E14.5, tooth development in Msx1^+/−-Msx2^−/− embryos exhibits no phenotype and resembles that in Msx1^+/−-Msx2^+/− and wild-type control embryos (Fig. 4A,B). Thus, prior to E14.5, Msx2 appears to play a non-essential role in tooth development. In Msx1^−/−-Msx2^+/− embryos, tooth development arrests at the bud stage, similar to the arrest observed in Msx1 single mutants (Fig. 4C) (Satokata and Maas, 1994). In contrast, molar tooth development in some Msx1^−/−-Msx2^−/− double mutant embryos arrests at the dental lamina stage, with only a rudimentary thickening of the oral ectoderm (Fig. 4D). In other cases, immature dysplastic tooth buds were observed (data not shown). Bmp4 and Fgf8 expression is preserved in the Msx1^−/−-Msx2^−/− double mutant dental lamina (data not shown). These results indicate that Msx1 and Msx2 function redundantly during tooth initiation in dental mesenchyme, where Msx2 expression is superimposed transiently upon that of Msx1 (MacKenzie et al., 1992). Moreover, the initial requirement for Msx gene function during tooth development coincides with the time when Dlx1 or Dlx2 gene function is required in prospective maxillary molar tooth development.

For Msx and Dlx to act in parallel in early tooth morphogenesis predicts that their mesenchymal expression should be coordinately regulated by dental epithelial factors. To determine whether FGF8 or BMP4 can induce Dlx1 expression in the dental mesenchyme, beads containing FGF8 or BMP4 were implanted in wild-type E11.5 dental mesenchymes. Beads soaked in BMP4 produced minimal or no induction of Dlx1 expression in the dental mesenchyme (data not shown, Table 1), while similarly treated beads strongly induced Msx1 and Dlx2 (see below). In contrast, FGF8 strongly induced Dlx1 expression in wild-type dental mesenchyme at E11.5 (Fig. 5A, Table 1). Control experiments employing beads soaked in BSA or using a Dlx1 sense riboprobe gave no signal (Fig. 5B, Table 1). Experiments combining wild-type dental epithelium with wild-type dental mesenchyme also showed induction of Dlx1 expression in the dental mesenchyme (arrow in Fig. 5A).

To determine whether BMP4 or FGFs induce Dlx2 expression in dental mesenchyme and whether this induction requires Msx1, beads containing BMP4, bFGF or FGF8 were implanted in isolated wild-type and Msx1 mutant E11.5 dental mesenchymes. BMP4 induced Dlx2 expression strongly and equally well in wild-type and Msx1 mutant molar mesenchymes (Fig. 6A,B, Table 1). Control experiments employing beads soaked in BSA or using a Dlx2 sense riboprobe gave no signal (Table 1). Induction of Dlx2 expression in the dental mesenchyme was also observed when wild-type dental epithelium was recombined with wild-type dental mesenchyme (data not shown). FGF8 also induced Dlx2 expression in wild-type dental mesenchyme and in Msx1 mutant molar mesenchyme (Fig. 6C,D). Control experiments employing beads soaked in BSA gave no signal (data not shown and Table 1). Thus, Msx1 is not required for Dlx2 induction in the dental mesenchyme by either BMP4 or FGF8.

Based on the above results, a genetic model for early tooth development is presented (Fig. 7). At the dental lamina-early bud stage (E11.5-12.5), the model places epithelial Fgf8 upstream of mesenchymal Msx1 for the following reasons. Fgf8 is known to be expressed as early as E10 in the oral ectoderm (Neubüser et al., 1997, and refs. cited therein), FGF8 is able to induce Msx1 expression in the dental mesenchyme, mimicking a function of the oral ectoderm (Kettunen and Thesleff, 1998) and epithelial Fgf8 expression is preserved in both Msx1 and Msx1,Msx2-deficient dental epithelia. Beginning at E12.5, Fgf3 is expressed in dental mesenchyme (Thesleff and Vaahtokari, 1992), and Msx1 is placed upstream of mesenchymal Fgf3 because Fgf3 expression is not detected in Msx1 mutant dental mesenchyme. This model is further supported by bead implantation experiments demonstrating that FGF8 and other FGFs can induce Fgf3 expression in the dental mesenchyme in a manner that requires Msx1. In addition, BMP4 is not able to induce Fgf3 expression in wild-type dental mesenchyme and FGFs are not able to induce Bmp4 expression (Chen et al., 1996). These results indicate that epithelial BMP4 and FGF8 act by different Msx1-dependent pathways to induce expression of members of their respective gene families in dental mesenchyme. Thus, the initiation step of tooth development may be considered as consisting of at least two separate Msx1-dependent pathways, corresponding to Msx1 induction by either FGF8 or BMP4 (Fig. 7).

FGF8 is known to act as an inductive signal in different developmental systems (Vogel et al., 1995; Lee et al., 1997; Crossley et al., 1996; Richman et al., 1997). Our data suggest that an FGF-dependent signaling pathway participates in the regulation of tooth morphogenesis. Previously, BMP4 has been proposed to act as a signaling molecule mediating early epithelial-mesenchymal interactions during tooth morphogenesis (Vainio et al., 1993). However, since FGF8 is able to induce Fgf3 expression in the dental mesenchyme at the lamina-bud transition stage when odontogenic potential shifts from the epithelium to the mesenchyme, FGFs may constitute an additional component of the signaling cascade mediating odontogenic epithelial-mesenchymal interactions.

For the moment, the role of mesenchymal FGFs such as FGF3 in odontogenesis remains unclear. Fgf3 knockout mice do not exhibit an overt tooth phenotype (Mansour et al., 1993) and FGF3 has been excluded as a gene defect in human hypodontia (Nieminen et al., 1996). However, this could be explained by functional redundancy between FGF3 and mesenchymally expressed FGF7 (Finch et al., 1995). FGF receptors are expressed during tooth development in both dental mesenchyme and epithelium. The FGFR2b splice variant, encoding a potential receptor for FGF3, FGF7 and FGF10, is expressed in the cap stage enamel organ, while FGFR1, capable of binding several FGFs, is expressed in dental papilla mesenchyme (Orr-Urtreger et al., 1991, 1993; Peters et al., 1992; Ornitz et al., 1996; Igarashi et al., 1998). It is thus possible that, analogous to the function proposed for mesenchymal BMP4 (Chen et al., 1996), mesenchymal FGFs bind to epithelial receptors to promote further epithelial development. A similar model where an epithelially expressed FGF ligand binds to a receptor that is expressed in adjacent mesenchyme and vice versa has been proposed for the limb bud (Xu et al., 1998).

It is also possible that FGF3 has direct effects within the dental mesenchyme. Since the FGF bead implantation experiments are performed at E11.5 but culture continues for an additional 24 hours, the induction of Msx1 expression by FGFs could mimic either the effects of epithelial FGF8 or the effects of endogenous mesenchymal FGF3. The model presented here is compatible with mesenchymal FGF3 helping to maintain Msx1 expression by a positive feedback loop. In addition, Fgf3 was originally isolated as the int-2 proto-oncogene, and thus is implicated in the regulation of cellular proliferation (Dickson and Peters, 1987). Other FGFs have been shown to induce cell proliferation in dental mesenchyme (Jernvall et al., 1994) and FGF3 could represent an endogenous FGF that performs this function. Further experiments are required to address whether mesenchymal FGF3 acts directly upon the dental epithelium or acts autonomously within the dental mesenchyme.

Mice homozygous deficient for both Dlx1 and Dlx2 genes exhibit an arrest of upper molar tooth development at the dental lamina stage, and in situ hybridization experiments show that epithelial Fgf8 and mesenchymal Msx1 and Bmp4 expression is preserved in the Dlx double mutant upper molar mesenchyme (Thomas et al., 1997). Conversely, in Msx1 mutants, we find that at the initiation stage, epithelial Fgf8 and mesenchymal Dlx1 and Dlx2 expression is preserved, while subsequent mesenchymal Bmp4 expression is absent. Moreover, BMP4 and FGF8 are able to induce both Msx1 and Dlx2 expression in the dental mesenchyme while Dlx1 expression is induced by FGF8. These results suggest that Msx and Dlx might act in parallel at the dental lamina stage (Fig. 7). Analysis of tooth development in Msx1,Msx2 double mutants further supports this idea. While the Msx1 mutant arrests at the bud stage and the Msx2 mutant exhibits only later defects in tooth development (M. B., unpublished data), Msx1,Msx2 double homozygotes exhibit an arrest at the dental lamina stage, which affects all molar teeth, but otherwise phenocopies the Dlx1,Dlx2 double mutant phenotype in the maxillary molar dentition. This suggests that Msx1 and Msx2 function redundantly in early dental mesenchyme, and indicates that Msx1 or Msx2 function is required as early as tooth initiation, despite the later bud stage arrest in Msx1 mutants. Furthermore, recombination experiments confirm that Dlx1,Dlx2 and Msx1 function is required solely in the dental mesenchyme for tooth morphogenesis (Thomas et al., 1997; K. Kratochwil, personal communication). Thus, at the initiation stage, we propose that Dlx1,Dlx2 and Msx1,Msx2 act in parallel in the dental mesenchyme, subject to differential induction by epithelial BMP4 and FGF8. Although these results do not show whether Msx or Dlx expression is directly activated by BMPs or FGFs in mouse tooth mesenchyme, in Xenopus Msx1 expression is an immediate early response to BMP4 (Suzuki et al., 1997).

In contrast to the pathway presented for the initiation stage of tooth development, at the bud stage, Dlx2 is placed downstream of mesenchymal Bmp4 because Dlx2 is reduced in Msx1 mutant dental mesenchyme and bead implantation experiments show that BMP4 is able to induce Dlx2 expression even in the absence of Msx1. These results suggest that, while Dlx1 and Dlx2 are likely to function in parallel with Msx1 and Msx2 at the lamina stage, Dlx2 expression at the bud stage resides downstream of Msx1. This latter relationship suggests a requirement for mesenchymal Fgf3 and Bmp4 to maintain Dlx2, but not that of Dlx1, in the dental mesenchyme. However, the data do not exclude the possibility that, at the bud stage, Dlx2 is a direct target for regulation by the Msx1 gene product. In Dlx1,Dlx2 mutants, an abnormal area of Barx-1-negative mesenchymal cells underlying the maxillary molar tooth germ coincides with an area of ectopic Sox9 expression and with ectopic cartilage formation (Thomas et al., 1997). This result has been interpreted as a patterning defect resulting from mis-specification of migratory or premigratory odontogenic neural crest, distinct from mutants in epithelial-mesenchymal inductive tissue interactions such as Msx1 and Lef1 (Thomas et al., 1997). We suggest two provisos to this interpretation of the Dlx1,Dlx2 mutant phenotype. First, the conclusion that the Dlx1,Dlx2 mutant is unique as a patterning mutant must acknowledge that the most likely basis for the selective loss of maxillary molars is redundant function with Dlx3, Dlx5 and Dlx6 which are expressed in mandibular but not maxillary mesenchyme (Thomas et al., 1997; Qiu et al., 1997). Thus, there is no particular reason why the Dlx and Msx mutants need to be mechanistically distinct. Second, it is by no means clear that similar findings might not also pertain to tooth development in Msx1,Msx2 double mutant embryos. For example, in the absence of proper induction of the dental mesenchyme, odontogenic mesenchyme might well be expected to acquire an altered fate such as cartilage or bone. In fact, newborn Msx1 mutants exhibit ectopic bone formation in the region normally fated to form the molar tooth (see Fig. 3f, Satokata and Maas, 1994). Thus, we favor the interpretation that similar to the Msx1,Msx2 mutant, the Dlx1,Dlx2 mutant represents an early inductive failure within the dental mesenchyme.

Factors such as BMP4 are able to induce Msx1 expression in dental mesenchyme, but they are not able to induce the expression of genes such as Fgf3 for which Msx1 function is required (Fig. 7). This indicates that Msx1 is necessary but not sufficient for the induction of these signaling molecules in dental mesenchyme, and suggests that additional, coordinately induced factors are needed for Msx1 to regulate gene expression. Thus, it is attractive to think that additional factors might interact directly with Msx1 or, alternatively, might act separately from but simultaneously with Msx1 on target gene promoters to regulate transcription. Recent in vitro data show that Msx1 and Dlx2 can form homodimeric and heterodimeric complexes via dimerization through their homeodomains (Zhang et al., 1997). Although it is unclear whether these homeoproteins interact or regulate transcription in vivo, a model has been proposed whereby Msx and Dlx gene products inhibit each others transcriptional properties (Zhang et al., 1997).

Other potential candidates for this role are members of the Lim-homeodomain family, Lhx6 and Lhx7, which are specifically expressed in the oral aspect of the maxillary and mandibular mesenchyme in response to FGF8 but not BMP4, and are subsequently expressed in tooth mesenchyme (Grigoriou et al., 1998). In addition, Pax9, a member of the Pax family, is strongly expressed in the dental mesenchyme from E12-14, and Pax9 knockout mice exhibit an arrest of tooth development that phenocopies the Msx1 mutant (Peters et al., 1998).

Interestingly, Pax9, like Msx, Dlx and Lhx is differentially regulated by BMP4 and FGFs (Neubüser et al., 1997). It has been shown that Bmp4 and Fgf8 exhibit wide but overlapping expression domains in the mandibular arch and that BMP2 and BMP4 antagonize the induction of mesenchymal Pax9 by FGF8, thus helping to establish the position of prospective tooth mesenchyme. Although our bead implantation experiments were not analyzed for such antagonistic effects, it is clear that FGF and BMP epithelial signals provide different molecular functions in induction of the dental mesenchyme. In contrast to its repressive effects on Pax9, BMP4 activates the mesenchymal expression of Msx1, Msx2 and Dlx2. Thus, the specific combination of factors that are induced in dental mesenchyme will differ depending upon whether the overlying ectoderm is expressing FGF8, BMP4 or both. The existence of spatially and molecularly distinct signaling pathways could provide a mechanism for the conversion of distinct ectodermal expression domains into unique combinations of transcription factors that are expressed in spatially distinct domains in the underlying mesenchyme. This would serve to preserve – as also occurs in the developing limb – the information that will provide polarity and pattern during subsequent organ development.

We wish to thank Drs John L. R. Rubenstein (UCSF) and Gail Martin (UCSF) for the Dlx and Fgf probes, respectively. We particularly wish to thank Dr Irma Thesleff (University of Helsinki) for her encouragement and for sharing unpublished information, and Drs Bjorn Olsen, Andy McMahon and Ichiro Nishimura for constructive comments. This work was supported by grants from the NIH (DE11987) to R. M. and from the American Association of Orthodontists to M. B.