Correlation of wing-leg identity in ectopic FGF-induced chimeric limbs with the differential expression of chick Tbx5 and Tbx4

In vertebrates, two pairs of limb buds are formed from the lateral plate mesoderm at specific anteroposterior positions and grow out to form the limbs. The morphology of the anterior limb, the forelimb, is quite diverse among vertebrates; from wings in birds to hands in primates. Although the difference between the forelimb and the posterior limb, the hindlimb, appears small in rodents, the hindlimb is usually larger than the forelimb as it serves to support the body, especially in avians and humans.

One of the important questions in phylogeny and developmental biology has been how morphological and functional differences among different limbs are determined. Interesting implications can be drawn from recent experiments using chick embryos in which ectopic limb formation is induced in the flank by application of members of the fibroblast growth factor (FGF) family (for a review; Cohn and Tickle, 1996). The extra limbs thus formed exhibited wing-like phenotypes when induced in the anterior flank, whereas those in the posterior flank resemble legs (Cohn et al., 1995; Ohuchi et al., 1995).

This suggests that determination of limb identity may depend on the axial level of the limb. Furthermore, Cohn et al. (1997) recently showed that activation of the expression of a certain set of Hox9 paralogous genes in the lateral plate mesoderm may be correlated with the specification of each limb.

In addition, several factors have been found to exist specifically in the forelimb or hindlimb region. For example, one of the Hox gene products, Hoxc-6 protein is specifically distributed in the forelimb region (Oliver et al., 1988; Nelson et al., 1996), whereas Hoxc-9, Hoxc-10 and Hoxc-11 are expressed in the hindlimb bud (Peterson et al., 1994, Nelson et al., 1996). However, the expression of these Hox proteins or RNA are further restricted within the limb bud, for example, to either the anterior or posterior region. Recently it has been shown that some T-box genes are expressed throughout the entire domain of either limb bud. The T-box genes belong to a family of putative transcription factors, characterized by a region of homology to the T (Brachyury) locus gene product (Herrmann et al., 1990; Bollag et al., 1994). It was found that the expression of two vertebrate T-box genes, Tbx5 and Tbx4 are predominantly restricted to the forelimb and hindlimb buds, respectively (Chapman et al., 1996; Gibson-Brown et al., 1996; Simon et al., 1997). Furthermore, TBX5 mutations have been reported to cause Holt-Oram syndrome, resulting in some defects of the upper limb (Basson et al., 1997; Li et al., 1997), implying that Tbx5 may be involved in forelimb development. Tbx5 and Tbx4 thus represent good marker genes for forelimb and hindlimb identity, respectively.

To understand the mechanisms of determination of limb identity, we analyzed the phenotypes of the additional limb induced by FGF in the flank. We found that the anterior half of an ectopic limb has characteristics of the forelimb, whereas hindlimb features are present in the posterior half, demonstrating the chimeric character of the ectopic limb. We used Tbx5 and Tbx4 as marker genes to examine the identity of the additional limbs. We found that the expression of both Tbx5 and Tbx4 are induced by FGF in the flank and that there exists a clear boundary between Tbx5/Tbx4-expressing and non-expressing domains in the extra limb bud. The Tbx expression boundary in the additional limb appears to correspond to its future line of demarcation between the wing and leg phenotypes. Thus, the Tbx5/Tbx4 expression appears correlated with the limb identity in the additional limb, suggesting that Tbx5/Tbx4 may be involved in determination of limb identity in vertebrates.

Two types of retroviruses were used for the present study, one is a replication-defective variant of the spleen necrosis virus (RDSNV; Dougherty and Temin, 1986), and the other a replication-competent variant of the avian leukemia virus with splice acceptor site, Bryan high titer polymerase, and A-envelope subgroup (RCASBP(A); Hughes et al., 1987). Construction of the RDSNV containing the human Fgf4 cDNA and propagation of the recombinant viruses and that of the RCASBP(A) containing the chick Fgf8 cDNA were described previously (Ohuchi et al., 1995, 1997a; Itoh et al., 1996).

Fertilized chick eggs were purchased from Yamagishi Co. (Tokushima, Japan) or from Nisseiken Co. (for specific pathogen free chicks, line M; Yamanashi, Japan) and incubated at 37.5-38.5°C until the desired stages (Hamburger and Hamilton, 1951). Chick embryonic fibroblasts (CEFs) were grown as described previously (Ohuchi et al., 1997b). For grafting experiments, CEFs were first infected with the viruses, cultured for 3 days, and harvested by a light trypsinization and centrifugation for 5 minutes at 800 revs/minute. The resulting cell pellet was then cut into pieces using tungsten needles. Before grafting, sterile black ink (Pelikan, 1:20 dilution) was injected below the blastoderm to visualize the embryo. Using fine tungsten needles, a small slit was made into the appropriate place of the lateral plate mesoderm and a piece of cell pellet was grafted into the prepared slit. The eggs were resealed and returned to the incubator. Embryos were incubated for further 48 hours to 11 days depending on the experiment.

Heparin acrylic beads (H5263, Sigma), 125-250 μm in diameter, were soaked in 10 μl of human recombinant FGF2 (1 mg/ml; kindly provided by Takeda Pharmaceutical Company) for 1 hour at room temperature prior to implantation. Just before application to the embryos, the FGF2 beads were transferred to tissue culture medium to avoid application of highly concentrated FGF2-drops. A bead was inserted into the slit in the lateral plate mesoderm at the appropriate anteroposterior position. Embryos were incubated for a further 7-11 days for cartilage staining or observation of cutaneous structures. To observe the epithelial patterns of extra limbs, the manipulated embryos were fixed in 4% formaldehyde at embryonic day 12-13. We confirmed that FGF2-soaked beads and Fgf-expressing cells have essentially the same effect on the flank cells in inducing an extra limb as reported previously (Crossley et al., 1996; Vogel et al., 1996; Ohuchi et al., 1997b).

After manipulation, embryos were allowed to develop for 7-9 days, then fixed in 4% paraformaldehyde and stained for cartilage using 0.1% alcian blue. After dehydrating and clearing with methyl salicylate, the cartilage structures of the embryos were observed with a Leica MZ-series microscope.

To isolate chick Tbx5 cDNA, mRNA was extracted from chick wing buds at embryonic day 3 using the QuickPrep Micro mRNA Purification Kit (Pharmacia Biotech). 2 μg mRNA was reverse transcribed using the Ready-To-Go T-Primed First-Strand Kit (Pharmacia Biotech). Polymerase chain reaction (PCR) was performed using degenerate oligonucleotide primers, which were derived from highly conserved regions within the T-box and correspond to the following peptide sequences: YVHPDSP for sense and AVTSYQN for antisense (Agulnik et al., 1996). PCR products were gel purified, cloned into pCR2.1 according to the manufacturer’s instructions (TA cloning Kit, Invitrogen Corporation). Sequencing analysis revealed that one of the plasmids contained a 252 bp fragment of chick Tbx5 cDNA. To obtain a longer Tbx5 clone, rapid amplification of cDNA ends (Frohman et al., 1988) was performed using the Marathon Kit (Clontech Laboratories, Inc.). The resultant 1.3 kb clone, cTbx5-3 was used for in situ hybridization analysis.

Chick Tbx4 cDNA was obtained by screening a chick limb bud (stage 20-23) cDNA library prepared in lambda ZAPII phage vector (Stratagene) with a human Tbx2 cDNA clone (a kind gift from Dr Christine E. Campbell). Among the 13 positive clones isolated was a clone for chick Tbx4 (cTbx4-N; 2.3 kb). Tbx4 RNA riboprobe was synthesized from cTbx4-N for in situ hybridization analysis.

Embryos were fixed in 4% paraformaldehyde 48 hours after surgical manipulation and prepared for whole-mount in situ hybridization, which was performed according to Wilkinson (1992) with minor modifications (Ohuchi et al, 1997b). Digoxigenin-UTP riboprobes were generated from the T-box sequence, coding sequences 3′ to the T-box, and 3′ untranslated sequences of cTbx4-N and cTbx5-3. To visualize the somites, a chick cDNA encoding a muscle transcription factor, MRF4 (1.1 kb; Fujisawa-Sehara et al., 1992) was used. According to Burke et al. (1995), the chick forelimb bud extends from somites 18-19 through somites 21-22 by stage 24. When Mrf4 was used for a somite marker, the posterior-most level of the forelimb bud was counted as somites 20-21 in the stage 22-24 embryos that we examined. Therefore the somite levels reported here can be calibrated, plus one somite, compared with the report by Burke et al (1995).

To identify feather buds, a probe for chick Sonic hedgehog (Shh) (1.3 kb; Ohuchi et al., 1997a) was used on 10-day additional limbs and contralateral wings and legs as reported previously (Nohno et al., 1995).

We designated the FGF-induced extra limb in the flank as Dasoku (snake leg); a Japanese word for a superfluous limb of the snake (Ohuchi et al., 1995). To analyze the patterns of cartilage and integument of these Dasokus, we implanted Fgf-expressing cells (FGF4-cells or FGF8-cells) or an FGF2-soaked bead in the prospective interlimb region of stage 13-15 chick embryos. At 7 days after implantation, we analyzed the cartilage pattern of the Dasokus. As reported previously (Cohn anteroposterior polarity, e.g., the sequence of digits in Dasokus read 4-3-2 from anterior to posterior, which is reversed compared with the sequence of normal wing digits 2-3-4. Furthermore, the skeletal pattern of Dasokus changes according to their positions along the anteroposterior body axis. We classified 68 cases of Dasokus into three types according to their skeletal patterns; wing-type is defined as a Dasoku articulating with an authentic forelimb skeletal element such as scapula or humerus (Fig. 1A,B), intermediate type as a Dasoku which does not articulate with a forelimb or hindlimb element, but with fused ribs (Fig. 1C,D), and leg-type as a Dasoku articulating with an authentic hindlimb element such as pelvis or femur (Fig. 1F,G,H). All the FGF members that we examined, FGF2, FGF4 and FGF8, were similarly effective in the induction of all three types of additional limbs. It is noticeable that when FGF-cells were applied in the flank at the level of somites 22-23 of stage 14-15 embryos (34 cases), all three types of Dasokus were formed (Table 1). A possible explanation may be that, because of the very rapid growth rate of the embryonic cells, it was too difficult for us to apply FGF to the embryos under the exact conditions every time. Since the competence of the lateral plate mesoderm to respond to the FGF signal may change along the body axis due to the cephalocaudal gradient, i.e., the competence is lost earlier in more anterior flank, FGF could therefore induce a more leg-like extra limb even if FGF was applied to the relatively anterior flank of a more developed embryo.

Cartilage staining revealed that a full range of skeletal elements from humerus to digits were formed in the wing-type Dasokus. Many exhibited duplication of the humerus and radius, and thickening of the ulna (Fig. 1B). In 3 out of 34 cases, the wing on the implanted side was truncated (Table 1; data not shown), possibly due to the action of FGF within the wing territory, as mentioned previously (Vogel et al., 1996). In the leg-type Dasokus, a full range of skeletal elements from pelvic girdle to digits were formed in which the ectopic femur articulated with an enlarged pelvic girdle (Fig. 1G,H). In the intermediate type Dasokus, skeletal elements from stylopods to autopods were formed, with the stylopods usually articulating directly with the fused ribs (Fig. 1D). In normal chick embryos, the ribs (which do not fuse) articulate directly with the sternum (data not shown). Occasionally, a small cartilage fragment was formed independently between two costal bones, which seemed to be an extra girdle (data not shown). In contrast to the 3 digits of the normal wing and the 4 digits of the normal leg, the intermediate-type Dasokus had 2 to 5 digits (Fig. 1E), whose patterns do not seem to resemble either those of the wing or the leg. Eleven other cases displayed a combination of these phenotypes, such as formation of an additional humerus, thickening of the ulna, bifurcation of the tibia and presence of 5 digits in the common autopod of the duplicated wing and leg (data not shown).

Since the forelimb (wing) and the hindlimb (leg) in chicks can be discriminated more effectively by cutaneous structures than with the skeletal patterns, we examined the epithelial patterns of additional limbs at embryonic days 10-13. The normal chick embryonic wing has feather buds on the dorsal side of the distal autopod region in addition to the more proximal region (Fig. 2A). By contrast, no feather buds are present in the autopod region of the normal leg (Fig. 2B,C). The distal autopod region of the leg has several characteristics not found in the wing; scales, claws, ventral phalangeal pads and a ventral curvature of the toe region (Fig. 2B). In addition to the morphological analyses of epithelial structures in ectopic 13-day limbs, we examined ectopic 10-day limbs for feather bud formation by use of a molecular marker, Shh (Fig. 2I), since Shh has been reported to be expressed in the feather bud epithelium (Nohno et al., 1995; Ting-Berreth and Chuong, 1996).

In the intermediate-type Dasokus formed in the mid-flank, the anterior half nearest the authentic wing had feather buds, whereas the posterior half near to the authentic leg had scales and claws (Fig. 2E,F). Interestingly, we found that even the wing-type Dasokus had phalangeal pad-like structures in their caudal margins, which are absent from feather buds (Fig. 2D) and the leg-type Dasokus had feather buds in their cephalic distal margins (Fig. 2G-I). In most cases, these feather bud and leg structure domains appeared to be separated by a boundary (Fig. 2D,F,H,I). It is noteworthy that the region of this boundary occasionally had characteristics of both wings and legs: there were feather buds in the autopod region and a claw in the distal tip of the digit (Fig. 2H). Although chimerism was not so obvious in the wing- or leg-type Dasokus by macroscopic observations of cartilage patterns, these epithelial characteristics of additional limbs indicated that the ectopic limb may be essentially a chimera of the forelimb and hindlimb.

To further verify their chimeric character, we wished to examine the expression patterns of wing- or leg-specific genes in the Dasoku limb buds. It has been reported that two members of the T-box gene family, Tbx5 and Tbx4, are expressed in the mouse forelimb and hindlimb buds, respectively (Chapman et al., 1996; Gibson-Brown et al., 1996), so we decided to use these Tbx genes as wing- and leg-specific markers. To obtain the chick Tbx genes, we employed RT-PCR using degenerate oligonucleotide primers in addition to cDNA library screening. Fig. 3 shows the deduced amino acid sequences of the chick Tbx4 and 5 T-box region, aligned with those of the mouse homologues. Within this conserved region, the amino acid identity for each gene compared with the corresponding mouse homologue is 99% for Tbx5 and Tbx4.

Prior to analysis of their expression in ectopically induced limb buds, we checked the mRNA localization of the Tbx5 and Tbx4 genes during normal chick limb development. At stage 11, the earliest stage examined, Tbx5 was expressed in the developing heart but not in the lateral plate mesoderm, while Tbx4 expression was not yet detectable (data not shown). By stage 12/13, Tbx5 expression became detectable in the lateral plate mesoderm at and below the level of the heart, and in the prospective wing mesoderm (Fig. 4A). The onset of Tbx4 expression began slightly later, at stage 13, in the caudal-most region of the lateral plate mesoderm (data not shown; Fig. 3F at stage 14). By stages 14-15, before demarcation of the wing and leg bud from the lateral plate mesoderm, high levels of Tbx5 and Tbx4 expression can be seen in the prospective wing and leg mesoderm, respectively (Fig. 4B,C,G). At stages 17-18, when the limb buds have begun bulging outwards, these two Tbx genes are differentially expressed in the wing and leg buds, respectively (Fig. 4D,H). Transverse sections of embryos stained by whole-mount situ hybridization revealed that neither Tbx5 nor Tbx4 are expressed in the limb bud epithelium (data not shown). In contrast with the reported sustained expression of Tbx4 in mouse tail mesenchyme (Chapman et al., 1996), expression of chick Tbx4 in the tail bud significantly decreased from stage 18 onwards (Fig. 4H-J). This is interesting from the view of comparable biology, because chickens lack tails whereas mice have quite prominent tails. At stage 23, Tbx5 and Tbx4 are still specifically expressed in the prominent wing and leg buds, respectively (Fig. 4E,I,J), but not in the apical ectodermal ridge (AER) (Fig. 4E,J), as reported in the mouse (Gibson-Brown et al., 1996). In 4-day embryos, Tbx5 expression was also present in the flank mesoderm at and anterior to the level of somite 25-26 (Fig. 4E). In contrast, Tbx4 expression was restricted to the mesenchyme in the vicinity of the leg bud at this stage (Fig. 4J).

As has been observed in mouse embryos (Chapman et al., 1996), chick Tbx5 is also expressed in the developing eye and the mesenchyme surrounding the foregut, lung buds and trachea (Fig. 4A-E), and Tbx4 in the lung and tracheal mesenchyme (Fig. 4I).

Since the normal expression patterns of Tbx5 and Tbx4 in chick embryos correlate well with those observed in mice, it appears that these Tbx genes can function as good wing- or leg-specific markers. We next examined their expression patterns in three types of FGF-induced extra limb buds. Tbx5 was expressed in the cephalic half of the intermediate-type Dasoku limb bud, while Tbx4 was expressed in its caudal half (Fig. 5B,E). Although we had speculated that Tbx4 would not be expressed in the wing-type Dasoku and Tbx5 absent from the leg-type, we observed Tbx4 expression in the caudal margin of the wing-type Dasoku, in which Tbx5 was predominantly expressed (Fig. 5A,D). Correspondingly, Tbx5 was also expressed in the cephalic margin of the leg-type Dasoku, which otherwise almost entirely expressed Tbx4 (Fig. 5C,F). These Tbx5/Tbx4 expression patterns imply that the cephalic region of the additional limb identifies itself as a forelimb, but its caudal region as a hindlimb, and that all of these FGF-induced extra limbs in the flank may be more or less chimeric even those formed near the authentic limb. In addition, this result indicates that FGF can induce or maintain expression of Tbx5 and Tbx4 genes in the flank mesenchymal cells, which are usually not expressed in the flank at such high levels (Fig. 4E,I). Since the differential expression patterns of Tbx5 and Tbx4 in the additional limb buds correlates well with the chimeric characteristics of their future morphology, it is likely that Tbx5 and 4 genes are involved in determination of chick limb identity.

To examine whether the somite level of the phenotypic boundary built by FGF application is fixed or variable, we performed in situ hybridization on Dasoku embryos using Tbx5 and Tbx4 RNA probes, together with a myotome marker gene, Mrf4, to mark the somites (Fujisawa-Sehara et al., 1992). We could thereby determine the level of the somite opposite the boundary in Dasoku limb buds. Typical results are shown in Fig. 6, where the boundary is seen at somite 23 (Fig. 6B), 23/24 (Fig. 6D) or 25/26 (Fig. 6F), indicating that the forelimb-hindlimb boundary is not fixed at a particular somite level. We also defined the position of the additional limb bud as the somite level corresponding to the middle of the additional limb bud and examined its correlation with the level of the boundary. As shown in Fig. 7, the extent of the expression domains of Tbx5 and Tbx4 genes are limited; Tbx5 can be expressed in the authentic forelimb region through the flank region to somite 26, whereas Tbx4 can be expressed in the flank region from somite 22 through the leg region. We found that the somite level of the boundary varies depending upon the position of the additional limb bud; when the extra limb is formed in more anterior flank, the boundary exists at a more anterior somite level, whereas when the additional limb is established in the more posterior flank, the boundary exists at more posterior somite levels (Fig. 7). The boundary of the Tbx5 expression (open circles in Fig. 7) differs slightly from that of Tbx4 expression (closed circles in Fig. 7). This result implies that the expression domain of Tbx5 can slightly overlap with that of Tbx4. This is consistent with the fact that the wing-leg boundary regions of Dasokus occasionally exhibit a mixture of both limb phenotypes (Fig. 2H).

Since the posterior end of the chick wing bud is at the level of somite 21-22 and the anterior end of the leg bud is at somite 29-30 by stage 24 (Burke et al., 1995), our results indicate that the somite level of the FGF-induced forelimb-hindlimb boundary can be altered within the interlimb region between the level of somites 22-26 (Fig. 7). This suggests that FGF signaling is required for Tbx5 and Tbx4 to be expressed in the prospective limb regions, but the expression domains of each Tbx gene are further determined by additional factors.

We demonstrated that three types of skeletal patterns of additional limbs, wing-type, leg-type and intermediate-type can be formed by ectopic application of FGF in the chick embryonic flank. Although the wing-leg phenotypic boundary is not clear in skeletal elements, most of the FGF-induced limbs were found to be chimeric in integumental characteristics and to exhibit a clear boundary in the integumental pattern. We also showed that the anterior half of the additional limb expresses Tbx5, whereas the posterior half expresses Tbx4 in the intermediate type. Since the expression domains of Tbx5 and Tbx4 in the extra limb bud seem to correspond to the future cephalic wing-like limb and the caudal leg-like limb, respectively, concomitant expression of Tbx5 and Tbx4 may be involved in forelimb and hindlimb chimerism.

The expression of Tbx5/Tbx4 genes could be closely correlated to establishment of the ectodermal structures rather than that of limb-type-specific skeletal features, because chimerism is not so obvious in cartilaginous elements (Table 1; Fig. 1), as observed in the pattern of epithelial structures (Fig. 2) and the pattern of Tbx5/Tbx4 expression (Fig. 5) in additional limbs. However, the skeletal patterns of the additional limbs are clearly different from those of the wing and leg, especially in intermediate-type Dasokus (Fig. 1), although it is difficult to verify their chimerism. If there were any appropriate molecular markers for limb-type-specific cartilaginous elements, one might be able to clarify whether Tbx5/Tbx4 specifies limb-type skeletal pattern. This idea is supported by the fact that skeletal abnormalities affect the upper limbs in Holt-Oram syndrome, caused by mutations in the human TBX5 gene (Basson et al., 1997; Li et al., 1997). Thus, Tbx5/Tbx4 may also be involved in specification of the skeletal pattern. Taken together with the expression pattern in normal chick (this study) and mouse embryos (Gibson-Brown et al., 1996), Tbx5 in the forelimb and Tbx4 in the hindlimb are most likely to be involved in determination of normal limb identity.

We found that a boundary exists between the Tbx5/Tbx4-expressing and non-expressing domains in the additional limb bud in the flank. In normal chick embryos by stage 24, the most posterior end of Tbx5 expression domain in the flank is at somite level 25/26 (Fig. 4E). This axial level corresponds to the thoracolumbar transition site of chick vertebrae by stage 25 (Burke et al., 1995). Therefore, we expected that the axial level of the Tbx expression boundary would be fixed at the level of the thoracolumbar transition even in the induced flank limb bud. However, we found that the level of this boundary varies with the site of FGF application between the levels of somites 22 and 26, not necessarily coinciding with the thoracolumbar boundary in the paraxial mesoderm. Thus, it appears that FGF activation of Tbx5 and Tbx4 gene transcription in the lateral plate mesoderm occurs independently of influence from the paraxial mesoderm. Furthermore, this implies that the position of the forelimb and hindlimb in tetrapods may be determined according to the site of FGF action in the lateral plate mesoderm. Recently, Cohn et al. (1997) showed that FGF can induce particular combinations of Hox9 paralogous gene expression in the lateral plate mesoderm reflecting the types of additional limbs formed in the flank. Therefore, it is likely that a Hox code may also constitute this FGF-Tbx signaling cascade.

Correlation of FGF with activation of the Tbx5 and Tbx4 genes is reminiscent of the upregulation of expression of other T-box-containing genes, such as TbxT and Tbx6L, by FGF4 (Knezevic et al., 1997). Furthermore, another T-box-containing gene Brachyury and embryonic FGF (eFGF) were shown to be components of the same regulatory loop in Xenopus (Schulte-Merker and Smith, 1995). Although T-box-containing genes are quite divergent outside of the T-box, it is possible that similar reciprocal interactions might be present between FGF and Tbx5/Tbx4 cognates in vertebrate limb specification.

Pinot (1970) reported that the prospective limb mesoderm can autonomously form a limb and determine its own limb identity depending upon its origin, but the prospective flank mesoderm cannot form a limb by itself nor specify limb identity. This phenomenon may be explained as follows: the prospective limb mesodermal cells can induce a limb bud in the presence of a mesenchymal FGF, such as FGF10 (Ohuchi et al., 1997), and determine limb identity by FGF-activated Tbx5/Tbx4 expression. By contrast, the flank mesodermal cells could form a Dasoku if they were supplied with FGF, thus inducing the concomitant expression of Tbx5 and Tbx4.

Classical experiments demonstrated that transplantation of leg mesoderm underneath the wing AER produces a leg distal structure in the host wing (Saunders et al., 1959). This implied that a secreted molecule(s) derived from limb mesoderm may be directly responsible for limb epithelial specification (for review see Gilbert, 1994). We showed that the mesenchymal expression patterns of Tbx5/Tbx4 closely correlate to its epithelial pattern: cephalic wing epithelium (feather buds) and caudal leg epithelial structures (scales, claws, etc). This observation suggests that the forelimb and hindlimb epithelial phenotypes may be specified by mesenchymal factors induced by Tbx5/Tbx4. Therefore, it is likely that Tbx5 and Tbx4 regulate particular downstream secreting factors to specify limb epithelium. Recently, it was shown that bone morphogenetic protein (BMP) signaling is required for scale formation in chicks (Zou and Niswander, 1996). In addition, a Tbx homologous gene, optomotor-blind was found to be regulated by decapentaplegic (a Bmp homologue) in Drosophila (Grimm and Pflugfelder; 1996). Since such homologous gene sets seem to be used in many developmental processes in insects and vertebrates, it is reasonable to predict Bmp as a candidate target gene for Tbx4. It has also been shown that FGF signaling is involved in feather bud development in chicks (Noji et al., 1993; Song et al., 1996; Widelitz et al., 1996). Therefore, FGF stands as another candidate target factor of Tbx5, at least in the case of wing epithelial specification. Since there exists a clear boundary between wing- and leg-epithelial phenotypes and between Tbx-expressing domain and non-expressing domain in additional limbs, we speculate that putative hindlimb and forelimb factors may suppress expression of Tbx5 and Tbx4, respectively, as observed in the case of Radical fringe and En-1 in establishment of the dorsal-ventral boundary during the AER formation (Rodriguez-Esteban et al., 1997; Laufer et al., 1997).

We propose a model for vertebrate limb specification as shown in Fig. 8. The present study demonstrated that FGF is involved in activation of Tbx5 and Tbx4, which then appear to function in determination of the forelimb and hindlimb identities, respectively. Mutual interaction between FGF and the Tbxs may be indispensable for initial limb specification, as has been shown in the case of Brachyury and eFGF in Xenopus (Schulte-Merker and Smith, 1995). Recent work has indicated that an initial endogenous FGF is most likely to be FGF10 (Ohuchi et al., 1997); however, as the main targets of FGF10 appear to be epithelial cells (Ohuchi et al., 1997), the effects of FGF10 on the mesenchymal cells might be indirect. FGF8 and FGF4, whose mRNAs are detected in the AER (Crossley et al., 1996; Niswander et al., 1993), seem to be dispensable for Tbx5/Tbx4 expression, because AER removal did not affect Tbx expression (H. Ohuchi, H. Yoshioka and S. Noji, unpublished results). Since FGF can upregulate expression of both Tbx5 and Tbx4, it would appear that FGF does not determine limb identity by itself. However, it has been shown that FGF can regulate the expression of specific Hox9 paralogous genes in the lateral plate mesoderm depending upon the somite level of FGF action (Cohn et al., 1997). This suggests that, in authentic limb formation, endogenous FGFs may determine the position of the limb by regulating the expression of Hox genes. Thus, expression of Tbx5/Tbx4 should be regulated at least by both FGF and Hox9 paralogous gene products. It has been found that a homeobox gene, P-otx/Ptx1 is specifically expressed in the hindlimb (Szeto et al., 1996; Lanctôt et al., 1997) and preliminary data reveals that it can be induced by FGF in the leg-type additional limb bud (H. Ohuchi, H. Yoshioka and S. Noji, unpublished results). Therefore, in determination of hindlimb identity, P-otx/Ptx1 may be involved in addition to the Hox genes and FGF, suggesting that these factors are likely to work cooperatively with Tbx5/Tbx4 in vertebrate limb specification. Subsequently, Tbx5 and Tbx4 activate their own sets of target genes to establish individual phenotypes with each limb identity. In determination of bone patterning, some factors may regulate Hox9-Hox13 codes for establishing cartilage pattern formation. For determination of epithelial phenotypes, mesenchymal cells expressing Tbx5/Tbx4 may produce some secreting factors such as BMP and FGF which act on epithelial cells. The consequent epithelial-mesenchymal interaction would result in epithelial pattern formation.

We thank Dr Ken Muneoka for comments on the manuscript, Dr Ruth Yu for critical reading of the manuscript, Takeshi Ohata, Hiroyuki Itoh, Ken Matsumoto, Masanori Takahashi, and Kiyomi Yoshida for technical assistance. We also thank Dr Yoichi Nabeshima for the chick Mrf4 probe, Dr Christine E. Campbell for human Tbx2 and Takeda Pharmaceutical Company for human FGF2 protein. This work was partly supported by grants from the Ministry of Education, Science and Culture of Japan, from The Mitsubishi Foundation and from ‘Research for the Future Program’, and by Special Coordinate Funds for Promoting Science and Technology from Science and Technology Agency of Japan to S. N.