Homeobox code model of heterodont tooth in mammals revised

Heterodonty is one of the hallmarks of mammals. It has been suggested that, homeobox genes, differentially expressed in the ectomesenchyme of the jaw primordium along the distal-proximal axis, would determine the tooth classes (homeobox code model) based on mouse studies. Because mouse has highly specialized tooth pattern lacking canine and premolars (dental formula: 1003/1003, for upper and lower jaws, respectively), it is unclear if the suggested model could be applied for mammals with all tooth classes, including human. We thus compared the homeobox code gene expressions in various mammals, such as opossum (5134/4134), ferret (3131/3132), as well as mouse. We found that Msx1 and BarX1 expression domains in the jaw primordium of the opossum and ferret embryos show a large overlap, but such overlap is small in mouse. Detailed analyses of gene expressions and subsequent morphogenesis of tooth germ in the opossum indicated that the Msx1/BarX1 double-positive domain will correspond to the premolar region, and Alx3-negative/Msx1-positive/BarX1-negative domain will correspond to canine. This study therefore provides a significant update of the homeobox code model in the mammalian heterodonty. We also show that the modulation of FGF-mediated Msx1 activation contributes to the variation in the proximal Msx1 expression among species.

previous study of BarX1 expression in the mesenchyme of multi-cusp tooth primordia 16 indicates the initial code might be inherited in the tooth germ mesenchyme. The homeobox code model had been proposed largely based on the data obtained from mouse studies, despite the fact that the mouse has a highly specialized dentition, possessing only 1 incisor and 3 molars for each quadrant of the jaw, separated by a diastema, a tooth-less region. Thus, it was unclear if this model could be directly applicable to other mammals, particularly to the species with all prototypical tooth types such as human. There are a few studies describing tooth development in species with all tooth types, such as ferret (Mustela putorius furo) [17][18][19] , house shrew (Suncus murinus) 20,21 , and gray short-tailed opossum (Monodelphis domestica) 22 . There are also some studies examining the expression of homeobox genes in various mammalian species 23,24 , but close comparisons of the homeobox code in mammalian species has not been done.
In this study, we have compared the expression of homeobox code transcription factors in the jaw primordium of various mammalian species, including mouse, ferret, and opossum, and found a large Msx1/BarX1-double-positive domain in the middle of the jaw primordium of the ferret and opossum embryos. As previously revealed, in contrast, such domain is small in the mouse mandibular arch. Histological analyses on various developmental stages of opossum suggest that the Msx1/BarX1-double-positive domain may correspond to premolars. Close observations of the homeobox expressions in both upper and lower jaws indicate that the canine would correspond to Alx3-negative/Msx1-positive/BarX1-negative region. Our explant culture studies indicate that the proximal expression of Msx1 in the mandibular arch is activated by FGF signaling. Thus, we suggest a revision of the homeobox code model in mammalian dentition. We also suggest that the diastema in mouse upper and lower jaws might be achieved, in part, by distinct mechanisms. Japanese quail (Coturnix japonica) eggs were obtained from local farms (Quail Factory-Tydess). Pregnant B6 mice were purchased from CLEA Japan, and midday of the vaginal plug was designated as embryonic day 0.5 (E0.5). The breeding colony of gray short-tailed opossum, is maintained at Nihon University School of Dentistry at Matsudo, under the approval of Nihon University Animal Care and Use Committee (No. AP12MD015, AP14MD014). Opossum embryos were obtained from pregnant females, with mating video-recorded and visually confirmed. Ferrets were maintained at Kanazawa University under the approval of Kanazawa University Animal Care and Use Committee as described previously 25,26 . Ferret embryos were obtained from pregnant female ferrets. The Committee for Animal Experiment of Tohoku University Graduate School of Medicine approved the experimental procedures in this study.

Materials and Methods
In situ hybridization. Whole-mount and section in situ hybridizations were performed as described previously 27 . Embryos were often bisected at the midline, and hybridized with distinct probes to accurately compare the expression of two genes at the identical developmental stage. The DNA fragments of opossum Alx3, Msx1, BarX1, Dlx1/2, and FGF8, corresponding to the exons were PCR-amplified from either oligo (dT) primed neonatal brain cDNA pool or genomic DNA prepared from M. domestica adult liver 28 . DNA fragments of ferret Msx1 and BarX1 were PCR-amplified from genomic DNA. The amplified DNA fragments were subcloned into pBlue-scriptII (Stratagene). The sequences of primers were listed in Supplementary Table 1 3D reconstruction. Postnatal opossums were decapitated and heads were fixed in 4% PFA/PBS. Brain and skin were removed, and the remaining jaw elements were embedded in paraffin. Serial sections were stained with hematoxylin and eosin with a standard method, and were photographed with KYENCE BZ-X710 microscope system. Tooth germs in the obtained images were manually filled, and the images were processed with Amira 5. 4. 3 (VSG Systems) for 3D reconstruction.
Explant culture. Explant culture of the mandibular arch and transplantation of drug-soaked bead were performed mostly as previously described 8 . Mandibular arch along with flanking tissues were dissected from E11.0 M. domestica, E10.0 mouse, and E3 quail embryos. One AG1-X2 (formate form) anion exchange bead (Bio Rad), soaked with either 10 mM SU5402 (197-16731, WAKO) in DMSO or DMSO alone, was transplanted into each mandibular arch. Explants were placed onto Millicell-CM culture plate insert (0.4um pore size, Millipore), and cultured in DMEM medium (Gibco) for 24 hours. Explants were subsequently fixed in 4% PFA/PBS, and processed for whole-mount in situ hybridization, as indicated above.
Correlation analysis of positions of tooth and mental foramen. Various mammalian skulls were digitally photographed and examined to correlate the position of mental foramen relative to the tooth along the proximal-distal axis of the dentary at Nihon University School of Dentistry at Matsudo (Chiba, Japan), the National Museum of Nature and Science (Tokyo, Japan), and the Ibaraki Nature Museum (Ibaraki, Japan). Specific specimen numbers are described in the figure legends and Supplementary Table 2.

Results
Homeobox gene expression and tooth classes in opossum jaw development. As opossum has all tooth classes (dental formula: 5134/4134, Fig. 1A), we considered it as a potential model to uncover the prototypical state of the "homeobox code" in mammalian dentition. Thus, we first examined the expression of the homeobox code transcription factors in the jaw primordium of opossum (Fig. 1C,D). At 11.5 days post-coitum (E11.5, Fig. 1B), when distal tips of left and right mandibular arches meet at the midline, distal-proximal-ordered expression of the "homeobox code" genes, such as Alx3, Msx1, BarX1, and Dlx2, was examined both in mandibular and maxillary primordium on whole-mount preparations (Fig. 1C). This pattern was largely similar to that in mouse mandibular arch at comparable stages, such as E10.5, as previously described 3,4 (See also Supplementary  Fig. 1). Interestingly, however, we have observed a large overlap of the proximal part of the Msx1 domain and most of the BarX1 domain in the opossum mandibular arch (Fig. 1C). This Msx1-BarX1 overlap was further confirmed on neighboring sections (Fig. 1C). In contrast, there is only a small overlap of expression domain of Msx1 and BarX1 in mouse mandibular arch 3,4 (See also Supplementary Fig. 1).
We hypothesized that the difference of the "homeobox code" gene expression between the mouse and opossum mandibular arches described above might reflect the difference in the dental formula of these species. Because lineage-tracing method of mandibular arch cells at distinct distal-proximal axis in the opossum embryos is currently not available, we cannot directly assess the correlation of the homeobox-coded domains and the fate of the mandibular cells in distinct domains. As the BarX1 expression in the proximal part of embryonic jaw primordia persists at later developmental stages in mouse 16,36 , we examined the expression of Msx1 and BarX1 at later stages of development to gain perspective on the spatiotemporal expression of Msx1 and BarX1 in the tooth germ. At E12.5 and E13.5 ( Fig Fig. 2). At E14.5 (post-natal day 0, P0), continuous dental lamina/tooth bands along the distal-proximal axis were observed both in upper and lower jaws, and the Shh-positive tooth epithelium is mostly at the bud stage (Fig. 2D). The most distal tooth germ mesenchyme expressed both Alx3 (not shown) and Msx1, but not BarX1 (Fig. 2D). In more proximal part of the tooth band, tooth germ mesenchyme expressed both Msx1 and BarX1 (Fig. 2D). Thus, although the later expression of Msx1 might be the result of local epithelial-mesenchymal interactions within the tooth germs 10-12 , the overlapped expression of Msx1 and BarX1 in the jaw primordium appeared to be maintained at least from E11.5 to P0 of opossum (Figs 1 and 2).
To correlate the tooth types and the expression domains of homeobox code genes, particularly Msx1 and BarX1, we further examined the morphology of the tooth band and tooth germs at later stages, such as P1, P2.5, and P10. For this purpose, we have made serial sections ( Fig. 3A-D), to generate 3D images of the tooth germs ( Fig. 3E-G, see also Supplemental Movie 1-3). At P1, we could observe sub-regionalization of the tooth band. Around the distal one third of the tooth band, the tooth germ epithelium at the bud stage appeared significantly taller (Fig. 3E, Supplementary Movie 1). The tooth germs located distally and proximally to this tall epithelium, were apparently smaller, but at the most proximal part of the tooth band, tooth germs were bigger at the early cap stage (Fig. 3B,E). This size tendency appeared intact at P2.5, although most tooth germs are at the late cap stage (Fig. 3C,F, Supplementary Movie 2). At P10, individual tooth germs were at the bell stage, and are morphologically distinguishable (Fig. 3D). Incisor region and premolar-molar region were clearly separated by canine ( The mandibular arch has no apparent morphological landmark along the proximal-distal axis to correlate the homeobox expressions with the tooth classes during early development (See Fig. 5, Supplementary Table 2, and the discussion for a potential landmark after jaw development completed). In contrast, upper jaw is made by a fusion of the medial/lateral nasal eminences and maxillary process, which give rise to premaxilla and maxilla, respectively. Thus, the boundary of these components could be used as a landmark to correlate the tooth types, the position of jaw primordium, and the homeobox expression. It is also known that the upper jaw incisors only grow from the premaxilla, and the canine grows at the distal end of the maxilla (Fig. 5, see also Figs 1A and 4A). Accordingly, when we observed the homeobox code expression in the upper jaw primordium, we realized that the homeobox expression was highly conserved between mouse ( Supplementary Fig. 1), and opossum (Fig. 1C). Alx3 expression in the upper jaw primordium was restricted to the nasal eminences. Msx1 expression was detected both in the nasal eminences and distal 2/3 of the maxillary process. BarX1 was expressed in the proximal 2/3 of the maxillary process, thus there is a significant overlap of Msx1 domain and BarX1 domain in the middle 1/3 of the maxillary process. Therefore, the future canine region likely corresponds to the Alx3-negative/Msx1-positive /BarX1-negative domain (Fig. 5). This observation is consistent with our inference that the premolar region will correspond to the Msx1/BarX1-double-positive domain (Fig. 5).
Msx1 and BarX1 expressions in ferret jaw primordium. The difference in the homeobox code expression in the mandibular arches of mouse and opossum embryos may simply reflect the difference of placentals and marsupials. We thus examined the expression of Msx1 and BarX1 in ferret embryos, because ferret is a placental mammal with all tooth classes (dental formula: 3131/3132, Fig. 4A). In E20 ferret embryos, in which left and right mandibular arches began to fuse at the midline, a large Msx1-BarX1-double-positive region was observed both in the maxillary process and the mandibular arch (Fig. 4B,C). We also observed the Msx1-positive/BarX 1-negative region at the distal part of the maxillary process, which will likely give rise to canine (Fig. 4B,C). It is of note that the distal limit of the BarX1 expression domain was close to the distal tip of the mandibular arch www.nature.com/scientificreports www.nature.com/scientificreports/ in the ferret embryos (Fig. 5), which might reflect the carnivore dentition (see also Discussion). These observations suggest that, in mammals with all tooth classes, the Msx1-positive/BarX1-negative domain and the Msx1/BarX1-double-positive domain will correspond to canine and premolar domains respectively, as indicated above (Fig. 5).

Regulation of homeobox expression in opossum mandibular primordium. It has extensively been
shown that, in mouse and chicken mandibular arch, while FGF8 expressed in the proximal mandibular arch epithelium of mouse embryos induces BarX1, BMP4 expressed in the distal mandibular arch epithelium induces www.nature.com/scientificreports www.nature.com/scientificreports/ Msx1 and represses BarX1 2,7-9,32 (See also Supplementary Fig. 1), contributing to the reciprocal expression of Msx1 and BarX1, with a small overlap (Fig. 6). If this is also the case in the opossum and ferret mandibular arch, how the large overlap of these genes could be possible? To obtain insights to answer this question, we have correlated the expressions of BMP4 and FGF8 with those of Msx1 and BarX1 in the mouse and opossum mandibular arches (Fig. 6). In the mouse mandibular arch, distal BMP4 and proximal FGF8 expressions well coincided with the distal Msx1 and the proximal BarX1 domains, respectively (Fig. 6, left panels). In fact, in the opossum mandibular arch, BMP4 and FGF8 were expressed in the distal and proximal epithelium, respectively (Fig. 6, right panels), very similar to those in mouse. While the BarX1 expression domain well coincided with the FGF8 expression in the opossum mandibular arch, however, Msx1 expression appears to be subdivided into weak distal tip expression and stronger proximal expression, and BMP4 expression only coincided with the distal Msx1-positive domain (Fig. 6, right panels). Thus, in the opossum mandibular arch, the proximal Msx1 expression might be activated by other mechanism(s). Since the proximal Msx1 expression domain in opossum appeared to overlap with the FGF8 domain (Fig. 6, right panels), and since exogenous FGF proteins could activate Msx1 expression in mouse mandibular arch explants 7,8 (See also Discussion section), FGF signaling could activate Msx1 expression in the opossum jaw primordium. Interestingly, when mandibular explants were prepared from mouse and quail embryos, with no implants or control DMSO-bead implantation, Msx1 expression could be detected all the way from distal tip to the proximal end of the explanted mandibular arch (Fig. 7A, Supplementary Fig. 3). This expanded expression of Msx1 was repressed by an inhibition of FGF signaling with a SU5402-bead implantation. Similar results were obtained with the opossum mandibular explants (Fig. 7B,C), indicating that FGF signal is generally required for the proximal Msx1 expression in the mandibular arch. Thus, in vivo, the proximal FGF-dependent Msx1 expression in the mandibular arch should be repressed by unknown mechanism(s) in mouse or quail, but in opossum, such repression might be either weaker or only effective in the very proximal part of the arch.

Discussion
In this study, we have compared the expression of the "homeobox code" transcription factors in the jaw primordia of mammalian species, and show that the expression pattern is well conserved in species with all 4 tooth-classes, such as opossum and ferret, but partially differs from that in mouse. The difference seems to be originated from the FGF-dependent activation of Msx1, leading to the broad overlap of BarX1 and Msx1, which will contribute to the premolar differentiation in opossum and ferret. We also indicate that the Alx3-negative, Msx1-positive, BarX1-negative domain seems to correspond to the canine. Accordingly, we suggest a revised version of the homeobox code model for mammals with all tooth classes (Fig. 5). This revised homeobox code fits well with a previous observation that the mesenchyme cells of tooth germs for multi-cusp tooth express BarX1 16 . This code had likely been established in the prototypical heterodont mammaliaformes/mammals, or at least in common ancestors of marsupial and placental mammals.
As most mammals including carnivores like ferret are diphyodont, initial "deciduous" teeth may be replaced with consecutive "permanent" teeth except for the molars 37 . In contrast, there is no tooth replacement in mouse. Thus, it should be carefully considered whether the homeobox code in the developing jaw would be involved in the fate determination of both sets, or only the initial set of the teeth. Interestingly, only the third premolars will be replaced with permanent teeth growing from the same dental lamina, and other initial teeth will be retained in marsupials including opossum 38,39 . Because the morphology of the second set of the teeth appears similar to that of the initial set, the second tooth germ may retain the initial homeobox code expression, or the memory of the code. establishment of mammalian homeobox code. Previous studies in mouse and chicken showed that the proximal-distal order of the homeobox code is largely conserved, thus the origin of the code precedes the separation of these lineages from the common ancestor. Accordingly, McCollum and Sharpe have suggested that www.nature.com/scientificreports www.nature.com/scientificreports/ the proximal-distal patterning of the jaw elements by differential homeobox expression established in a common ancestor of mammals and birds was thereafter co-opted for the heterodonty in the synapsid lineage 3 .
Yet, to achieve a divergent tooth patterns found in current mammals, the code would have likely been further modified. As indicated in the previous studies, and as confirmed in this study, distal Msx1 and proximal BarX1 domains have relatively small overlap in the mouse mandibular arch. In chicken embryos, these domains were previously reported as "complementary" 8 , like those in mouse embryos. However, when we looked at their data, we noticed a significant overlap of these domains, and we subsequently confirmed the overlap in quail embryos (Supplemental Fig. 3). Thus, as McCollum and Sharpe suggested, the prototypical homeobox code had been established in the common ancestor of mammals and birds, and it has been modified in mouse, particularly in the lower jaw. How modifications of the code might contribute to a variety of dental formula in mammalian species remain to be examined. It seems noteworthy that the distal limit of the BarX1 expression is very close to the distal www.nature.com/scientificreports www.nature.com/scientificreports/ tip of mandibular arch in ferret embryo, leaving a relatively small BarX1-negative domain. Because ferret, like other carnivora species, has canines in the front of the jaw, we speculate that such shift of BarX1 expression would contribute to the carnivorous adaptation.
It remains unclear how the proximal limit of the Msx1 expression is determined. As shown above, an isolation of the mandibular arch from the embryo provokes Msx1 expression in the entire mandibular arch taken from quail, mouse, and opossum. Thus, we hypothesize that inhibitory signal(s) would come from the rest of the embryonic body. This ectopic expression of Msx1 in culture is inhibited by an application of FGF signal inhibitor SU5402, suggesting the requirement of the FGF signal for the proximal Msx1 expression. In fact, some previous studies showed that, in the absence of the epithelium, exogenous FGF could induce Msx1 in the mouse mandibular explants 7,9 . Yet, while Msx1 expression nearby the FGF8 source is limited in the normal development, FGF signal itself likely affects more broadly, as other FGF targets, such as BarX1 2 and Lhx6/7 40 , are expressed more proximally. Interestingly, Msx1 activation by FGF in the chick limb bud is mediated by NF-κB 41 , suggesting that not the canonical MAP kinase pathway, but PI3 kinase/Akt or other pathway(s) could be used for the Msx1 expression. Thus, in the proximal part of the lower jaw primordia, this non-canonical FGF signaling may activate Msx1 expression, and this regulation may be further modified by unknown mechanism(s) at different degrees in various mammalian species. How co-expression of the Msx1 and BarX1 would affect the tooth morphogenesis is unknown, but a molecular interaction of Msx1 and BarX1 proteins 16 could regulate a unique set of target genes to achieve it.
Mouse upper VS lower tooth. Muroid rodents including mouse have a large diastema, separating remaining incisor and molars, while many rodent species such as squirrel possess one premolar adjacent to the first molar (see Supplementary Table 2) 42 . Previous mouse studies have shown that many rudimental tooth primordia are transiently formed in the future diastema of developing upper jaw, and such primordia degenerate afterwards by apoptosis [43][44][45] . In contrast, 2 rudimental tooth buds distally ranging the first molar is observed in the mouse lower jaw, and only epithelial thickening can be transiently recognized in the future diastema 43,44 . The difference of diastema formation, therefore, indicates that mechanisms of creating diastema in the upper and lower jaws may differ in mouse.
The mental foramen, an exit of trigeminal nerve from the dental bone, is located at the level between the first and the second premolars in human. We examined the position of the mental foramen in various mammalian species (Supplemental Table 2, See also Figs 1A and 4A). Although some species have multiple mental foramina, the major mental foramen locates at the level between the canine and the second premolar in most cases examined (Supplementary Table 2). Thus, it seems possible to use the mental foramen as a landmark of the canine-second premolar region in the dental bone, even if the given species may lack the canine and/or the premolar. In muroid rodents, in fact, the mental foramen is found close to the molars (Supplementary Table 2, See also Figs 5, 6, and Supplementary Fig. 1), suggesting that, in the mouse lower jaw, the premolar region may be largely missing, and the distal most of the diastema may simply be the socket of the rodent-specific huge incisor root. Consistently, a relatively small Msx1/BarX1 double-positive region, the premolar domain suggested in this paper (Figs 5 and 6), is observed in the mouse mandibular arch. Previous studies in mouse have shown that mutations involved in Fgf, Wnt, and Hedgehog signaling pathways result in an additional tooth in front of the first molar 12,14,15 . Such "revitalization" 46 of the rudimental tooth may be interpreted as an atavism, as extant rodents and possibly the common ancestors of muroid and extant rodents have one premolar in the same position (see also Supplemental Table 2). Nonetheless, the fact that no more premolars or canine has ever been observed in any mouse mutants further indicates that mouse mandibular arch has a potential to grow only one premolar. It will be an interesting question for the future studies if the mechanism predicted in rodents could also be applied in other mammals with premolar diastema, such as the marsupial group Diprotodontia, including wallaby and koala, as an example of convergent evolution. Human condition and homeobox code genes. Human genetic studies have identified familial tooth agenesis genes, such as Wnt10a, Pax9, and Msx1 47-52 . In Msx1-deficient patients, loss of 2nd premolar is the most prominent phenotype, suggesting that Msx1 is important for the human premolar development 47,51,52 . It is unclear, however, whether such hypodonty in the Msx1-deficiency is caused by the altered regional identity of the jaw, or the defective tooth development, because Msx1 is also expressed in the tooth germ at later developmental stages. To clarify such problem, we will have to manipulate Msx1 expression in the developing jaw primordium of mammalian species with all tooth classes at specific timing.

Data Availability
All data generated or analyzed during this study are included in this published article and its Supplementary Information.