Hoxd-12 differentially affects preaxial and postaxial chondrogenic branches in the limb and regulates Sonic hedgehog in a positive feedback loop

Vertebrate limb development has served as an excellent model for unravelling mechanisms of pattern formation during embryogenesis. Recent insights have been gained as to the molecular nature of some of the secreted signals that regulate growth and pattern along the anterior-posterior (AP), dorsoventral (DV) and proximodistal (PD) limb axes. Limb outgrowth occurs in a proximal-to-distal order, with branching and segmentation of more proximal cartilaginous anlagen to produce more complex distal structures, and is regulated by FGF signals from the overlying specialized ridge of ectoderm running along the DV edge of the bud, the apical ectodermal ridge or AER (reviewed by Hinchliffe and Johnson, 1980; and Cohn and Tickle, 1996). A functionally defined zone of polarizing activity (ZPA) located in the posterior limb bud mesenchyme regulates AP polarity (reviewed by Hinchliffe and Johnson, 1980), which is mediated by Sonic hedgehog (Shh). Anterior misexpression of Shh produces polarity reversals and consequent mirror-image digit duplications (Riddle et al., 1993; Chang et al., 1994). DV polarity is regulated by the limb ectoderm (reviewed by Hinchliffe and Johnson, 1980). Cross regulation of secreted signals along different axes coordinates growth and patterning (reviewed by Cohn and Tickle, 1996). Fgf4 expression is initially induced in posterior AER by Shh and subsequently FGF4 forms part of a positive feedback loop that stimulates Shh expression in posterior mesenchyme. Likewise, Wnt7a (a dorsal polarizing signal) together with FGF4, stimulates Shh expression.

While some of the potential downstream nuclear mediators of these signalling events have been identified, particularly multiple members of the homeodomain class of transcriptional regulators, how these genes function and interact to regulate limb development is less well understood, although clearly differential growth regulation is an important feature (see for example, Dolle et al., 1993; Yokouchi et al., 1995; Goff and Tabin, 1997). Along the AP axis, 5′ members of the Hoxd cluster are expressed in overlapping, nested, posterior and distal zones of the limb bud colinearly with their chromosomal order, and ZPA grafts induce duplicated Hoxd expression domains that correlate with subsequent skeletal duplications (reviewed by Izpisua-Belmonte and Duboule, 1992). By analogy with the key roles of homeobox genes in specification of segmental identity in Drosophila (see review by Krumlauf, 1994), these results suggested a Hox code in which AP positional identity would be specified by the combinatorial expression of different Hoxd genes along the limb AP axis. Proximodistal identity might be similarly regulated by the clustered 5′Hoxa genes, which are expressed in nested domains along the limb PD axis (Yokouchi et al., 1991). In fact, for specification of axial mesoderm in vertebrates, there is support for the operation of a ‘Hox code’ (reviewed by Krumlauf, 1994; see also Duboule, 1995). However, such simple models break down in the limb. While ectopic Hoxd-11 expression in chick embryo limb buds results in ‘posterior’ transformations (Morgan et al., 1992) and Hoxa-13 misexpression yields apparent ‘distal’ transformations (Yokouchi et al., 1995) that are compatible with a code, similar experiments with Hoxd-13 do not produce an analogous outcome (Goff and Tabin, 1997). Most notably, targeted disruption of several Hoxd and Hoxa genes has generally resulted in complex and sometimes subtle limb phenotypes affecting multiple skeletal elements that are not readily reconciled with Hox code models (eg. Dolle et al., 1993; Small and Potter, 1993; Davis and Capecchi, 1994, 1996; Favier et al., 1995; Fromental-Ramain et al., 1996; Kondo et al., 1996). These analyses have also revealed a high degree of functional redundancy and interaction between both linked Hox genes and paralogous (homologous) and non-paralogous genes in different Hox clusters (Davis and Capecchi, 1996; Davis et al., 1995; Favier et al., 1996; Fromental-Ramain et al., 1996; Kondo et al., 1996; Zakany and Duboule, 1996). In the case of Hoxd-12 in particular, loss-of-function results in minimal limb defects, and hints at potential function only begin to be revealed in the context of compound mutants with other Hox genes (Davis and Cappechi, 1996; Kondo et al., 1996). In fact, the high level of functional overlap and interaction between various Hox genes in the limb has raised speculation that this serves to expand limb size and morphology repertoire in a population, increasing plasticity and adaptability both during development and evolution (Duboule, 1994).

To gain further insight into the role of Hoxd-12 in limb development, we have used a well-characterized mouse Hoxb6 promoter (Schughart et al., 1991; Eid et al., 1993; Becker et al., 1996) to selectively drive Hoxd-12 transgene expression in the developing limb bud and lateral plate mesoderm of mouse embryos. Depending on location relative to the main limb axis, Hoxd-12 misexpression can either promote or inhibit formation and proliferation of chondrogenic condensations that give rise to limb skeletal elements. Furthermore, Hoxd-12 misexpression can activate ectopic Shh expression and produce mirror-image digit duplications. We propose that Hoxd-12 is normally part of a positive feedback loop within the posterior mesenchyme that reinforces polarizing signals during limb outgrowth. Such a role for Hoxd-12 in regulating Shh is further supported by the finding that misexpression of Hoxd-12 in several lateral plate derivatives phenocopies luxoid/luxate mouse mutants shown to have ectopic Shh signalling (see Chan et al., 1995; Masuya et al., 1995, 1997). In vitro binding and activation experiments suggest that Shh is a direct target of Hoxd-12. Hoxd-12 or Hoxd-11 (which shares some similarities with Hoxd-12 in gain-of-function phenotype) will activate endogenous Shh expression in limb mesenchymal cells in the presence of FGF, suggesting that certain 5′Hoxd genes may participate together in a positive feedback loop, in conjunction with FGF signals from the AER, to amplify the Shh signal in the posterior limb bud.

Routine isolation, cloning, labeling, blotting, PCR and sequencing procedures were performed using standard techniques (Sambrook et al., 1989). The nucleotide sequence of the mouse Shh EcoRI-XhoI promoter fragment has been deposited in GenBank (accession no. AF019387).

Expression constructs containing the chick Hoxd-12 coding sequence and its native 5′ untranslated region (5′UTR) translate very poorly in vivo (S. Mackem, unpublished observations). Therefore the 5′ UTR sequences were replaced with those from the RSV src 5′UTR, which supports efficient translation (Hughes et al., 1987). The coding sequence of Hoxd-12 5′ of the BglII site was replaced with a 23 bp NcoI-BglII oligonucleotide that converted the ATG to an NcoI site and changed the second amino acid from cysteine to glycine. This altered coding sequence, extending from NcoI to BamHI in the 3′UTR, was cloned into the NcoI and BamHI sites of the Cla12Nco vector containing the src 5′UTR (Hughes et al., 1987). SV40 virus late splice signals and poly(A) addition signals were added to this construct by PCR amplification. For the intron, a mini-intron containing late 16S donor and acceptor sites (including nucleotides 486-555 fused to 1411-1497 of the SV40 genome, in pOBCAT4 provided by C. C. Baker) was amplified flanked by BglII and SalI sites and introduced into the BamHI and SalI sites of Hoxd-12/Cla12Nco. The SV40 late poly(A) addition signal (from nucleotides 2545-2765 of the SV40 genome) was amplified flanked by SalI and HindIII sites and introduced into the same sites of the Hoxd-12/Cla12Nco construct. The final construct was inserted as a ClaI fragment downstream of a 3.6 kbp Hoxb-6 promoter (Schughart et al., 1991), and the transgene excised from plasmid sequences with BssHII for zygote injections.

The transgene was injected into either FVB/N or CD1 zygotes, as described by Hogan et al. (1994). Embryos were transferred to foster mothers and recovered after birth for establishing lines or immediate analysis, or at various stages of intrauterine development for analysis. Transgenicity of embryos was determined by DNA extraction from viscera (or heads for E10-13) and Southern analysis using either a 600 bp chick Hoxd-12 probe (as below) or the SV40 250 bp poly(A) signal fragment described above. Whole-mount in situ hybridizations and preparation of riboprobes was essentially as described by Conlon and Rossant (1992). For transgene detection, a 600 bp SacI probe from chick Hoxd-12 that does not cross hybridize with the murine Hoxd12 was used (Mackem and Mahon, 1991). A mouse Sonic hedgehog probe was provided by A. McMahon and mouse Hoxd-12 probe by D. Duboule. Skeletons were visualized by staining with either 0.1% Alcian green in acid-alcohol followed by alcohol destaining and clearing in methyl salicylate for some E13-E15 embryos; or with Alcian blue and alizarin red followed by alkaline hydrolysis and glycerol clearing for most embryos E14.5 and older (as described by Kessel and Gruss, 1991). Long-bone lengths were measured by micrometer and compared to nontransgenic siblings.

The 1 kbp Shh EcoRI-XhoI promoter fragment was ³²P-5′-end-labeled and incubated, in the presence of a 200-fold excess of nonspecific carrier Φ X DNA, with Hoxd-12/glutathione-s-transferase (GST) fusion protein (GEX-15, R. Hutson, S. Aguanno and S. Mackem, unpublished data) bound to glutathione sepharose. Bound beads were incubated and washed with PBS/0.1% Tween 20/5 mM DTT, digested with T7 Gene 6 exonuclease (50 units, USB) for 5 minutes using manufacturer’s conditions, and DNA released by phenol extraction followed by ethanol precipitation and analysis on 6% polyacrylamide-urea gels.

Early (72 hours incubated) chick limb buds were dissected into PBS and divided into anterior and posterior halves or used whole, and were trypsinized (0.25% trypsin/PBS) and washed in DMEM with 10% FCS. For transfections, the 700 bp PstI-XhoI mouse Shh promoter region was cloned from Bluescript as a PstI-KpnI fragment into a promoterless chloramphenicol acetyltransferase (CAT) reporter. Full-length Hoxd-12 with the sarc 5′UTR from Cla12Nco (see above) was cloned into the expression vector pSG5 (Stratagene). A Hoxd-12 construct containing the HSV-1 VP16 activation domain was generated by PCR amplification of sequences encoding VP16 amino acids 398-479 (provided by T. Kristie) and cloning in frame into the amino terminal end of full-length Hoxd-12 in Cla12Nco, just 3′ of the initiator ATG, followed by transfer into pSG5. Cells were transfected by electroporation (BioRad, 0.3 kV, 350 μ F, 5× 10⁶ cells in DMEM with 10%FCS) with from 5 to 10 μg of reporter and expression vector, plated onto 35 mm dishes and cultured for 36 to 48 hours at 38.5°C in DMEM with 10% FCS. Extracts were prepared and CAT assays performed using standard protocols (Sambrook et al., 1989). Hoxd12-expressing retrovirus was generated by inserting the Hoxd12/Cla12Nco construct into RCAS BP and transfecting chick embryo fibroblasts (Hughes et al., 1987), a Hoxd-11 RCAS BP construct was provided by C. Tabin and RCAS BP was used as a negative control. For infections, cells were plated at a density of 1.5× 10⁶ in 12-well plates and exposed to viral supernatants (about 10-fold concentrated final) for 2-3 hours, after which fresh media containing 300 ng/ml FGF4 (gift from Genetics Institute, Inc.) and 200 ng/ml heparin sulfate were added and incubation continued for 48 hours. Cells were harvested in RNAzol and northern analysis performed on 1.2% agarose-formaldehyde gels using a chick Shh probe (provided by R. Riddle) and a chick β-actin probe.

To misexpress Hoxd-12 selectively, a 3.6 kbp Hoxb-6 promoter was used to drive expression of the transgenic construct (diagrammed in Fig. 1A). This promoter, previously characterized extensively in transgenic mice (Schughart et al., 1991; Eid et al., 1993; Becker et al., 1996), directs expression specifically to the posterior lateral plate mesoderm including the limb buds.

In the limb buds, the Hoxb-6 promoter directs expression selectively to the posterior mesenchyme of the forelimb bud and throughout the mesenchyme of the hindlimb bud. The hindlimb expression later (E12-E12.5) splits into dual anterior and posterior domains (see Fig. 1C). A chick Hoxd-12 transgene was used since the chick and mouse proteins are very similar both in structure and expression, but the mouse coding sequence contains an ambiguity with respect to splicing within its coding region (Mackem and Mahon, 1991; Izpisua-Belmonte et al., 1991), and because use of the chick gene allowed transgene expression to be easily distinguished from endogenous mouse Hoxd-12. Since Hoxd-12 expression is normally restricted to the posterior and later to the posterior-distal limb bud mesenchyme (eg. Fig. 1B), ectopic expression of the Hoxd-12 transgene was expected in the anterior hindlimb and in the lateral plate mesoderm.

In initial experiments, two stillborn primary transgenics displayed a phenotype (described below), while several live born primary transgenics appeared entirely normal. Liveborn founders produced transgenic embryos with no detectable expression of the transgene, with a single exception (data not shown, discussed below). Therefore, assuming that high level transgene expression may be incompatible with postnatal survival, several additional primary transgenic embryos ranging from E16.5 up to newborn were analyzed. Among the liveborn founders, a single line (#2917) was identified with no phenotype but in which variable, weak Hoxd-12 expression was detectable in the lateral plate mesoderm and limb buds of transgenic embryos. Mating of this line to generate homozygous embryos substantially increased transgene expression and produced about 20% of offspring with phenotypic changes that were all very similar to those seen in abnormal primary transgenic embryos (described below). Thus it is unlikely that the line 2917 phenotype is related to transgene integration site. Sixteen consecutive litters from E13.5 up to newborn were analyzed (summarized in Table 1). About 80% of the expected number of homozygotes displayed one or more features of a characteristic phenotype (Tables 1-3). As expected, a subset of embryos from line 2917 matings showed consistent, easily detectable transgene expression in the lateral plate mesoderm and limb buds (Fig. 1D-F). Notably, variability in expression level was observed even between paired limb buds of the same embryo (not shown), and extension of transgenic Hoxd-12 expression to include the anterior edge of the distal forelimb bud was also seen (arrow in Fig. 1F). Occasional expression in the anterior distal forelimb has been previously observed in transgenic analyses of the Hoxb-6 promoter (K. Schughart, unpublished observations). Consistent with observed transgene expression, phenotypic changes were often not bilaterally symmetric and sometimes involved the distal forelimb (anterior digits), as well as the hindlimb and certain other lateral plate derivatives. The same phenotypic variability (between limbs within one embryo and between different embryos) was observed among primary transgenic embryos. No abnormalities were seen in non-transgenic embryos.

In both primary transgenics and in offspring of line 2917 matings, limb phenotypes generally involved only ‘anterior’ structures. Five out of six primary transgenics with a phenotype had abnormalities in either hindlimb and/or forelimb, and comparable hindlimb and/or forelimb abnormalities were seen in 26 out of 29 abnormal offspring from line 2917 matings (Table 2). In the hindlimb, the anterior autopod (hand/foot) was most frequently affected. Digital changes consisted of conversion of the anteriormost digit I (big toe) to a triphalangeal digit with longer metatarsal (similar to digit II or occasionally III in morphology) and/or anterior digit duplications (Fig. 2B,C,E,F; Table 2). Changes in the tarsal elements of the autopod were generally limited to the anterior tarsals in the distal row and again consisted of ‘posterior’ transformations of cuneiforme I to II or III, and/or duplications of cuneiforme I (Fig. 2I,J; Table 2). Fusions of the naviculare with cuneiforme I or II were also observed. The proximal row of tarsals (tibiale, talus, calcaneus) were either unaffected or mildly reduced in size. In the long bones of the hindlimb, abnormalities were seen in the tibia (anterior long bone), which was shortened to a variable degree (hemimelia, Fig. 2B,C; Table 2). Concomitant bowing of the fibula was interpreted as a secondary change related to failure of the tibia to elongate. The femur was relatively unaffected.

Occasional, usually unilateral abnormalities in the anterior-most digits of the forelimb were also observed; again consisting of digit I (thumb) conversions to a triphalangeal digit with longer metacarpal and/or anterior digit duplications (Fig. 2G; Table 2). In a single case, an extra digit-like element arose from the posterior pisiform (Table 2, ks10). Forelimb changes were restricted to the digits and were never observed in more proximal elements (carpals, long bones), consistent with the distally restricted Hoxd-12 transgene expression in the anterior forelimb (see Fig. 1F).

Abnormalities were generally not observed in posterior limb structures; hence overexpression of Hoxd-12 in domains where endogenous expression normally occurs did not produce any apparent phenotype. The pelvis is traditionally considered part of the hindlimb and is derived entirely from lateral plate mesoderm (Chevalier, 1977). Four out of six abnormal primary transgenics and an additional 30% of abnormal embryos examined from line 2917 matings displayed abnormalities of the pelvis. The most commonly affected component was the pubic bone (in 7/13 embryos), which was reduced (incomplete or absent superior ramus, Fig. 3F arrows), or shortened so that the pubic symphysis remained widely open (Fig. 3E,F; Table 2). A proportionately small pelvis, probably due to decreased growth of both pubis and ischium, was also observed (in 5/13 embryos; see Fig. 3E; Table 2). In a single case, nearly complete agenesis of the pelvis was observed, with only a small ilium remaining (Fig. 3G). Interestingly, the pubic bone is phylogenetically the most anterior component of the pelvis; the ilium subsequently rotates anteriorly, and the pubis and ischium more posteriorly (Hinchliffe and Johnson, 1980). Consequently, if this is reflected in mammalian ontogeny, then the most frequent pelvic abnormalities seen in the Hoxd-12 transgenics also seemed to selectively involve more anterior components.

Three of six primary transgenics with phenotypes and 62% of abnormal embryos examined from line 2917 matings displayed abnormalities in the sternum. The sternal phenotype ranged from mildly split posterior sternebrae to severe splitting and near total sternal agenesis (Fig. 3B-D; Table 3). The latter, more severe phenotypes would produce a ‘flail chest’ that would compromise respiration and account for the high frequency of neonatal mortality in transgenic animals with phenotypes. These sternal changes appeared to follow a posterior-to-anterior order in severity and so could represent a kind of ‘posterior transformation’ due to repression by Hoxd-12 expression, since the paired sternal bands (which arise in the lateral plate and later fuse in the midline) do not normally form at all in the posterior-most lateral plate (Chen, 1952a,b; Chevallier, 1977).

Abnormalities consistent with posterior transformations were also seen in the ribs, usually in conjunction with sternal changes (with two exceptions). These included both fusions of ribs prior to joining the sternum (normally seen only in more posterior ‘free’ ribs), and a reduction in the number of sternal articulating ribs from seven to six (Fig. 3B,D; Table 3). In a total of two transgenic animals, an eighth sternal rib was observed (usually considered an anterior transformation), and this was thought to represent a sporadic event since it occurred so infrequently and was also observed in non-transgenic littermates in the CD1 mouse strain used.

Although the ribs are derived from somitic mesoderm, the rib tips develop in close association with the lateral plate mesoderm and potentially could receive signals from this region (Huang et al., 1996), explaining the rib abnormalities. Transgene expression was never detected in somitic mesoderm and axial skeletal changes were not seen in primary transgenics or in mated line 2917 offspring. Vertebral numbers and morphologies were entirely normal with a single exception of 5 rather than 6 lumbar vertebrae in one animal, which was considered a sporadic event since it occurs as a natural variant in some mouse strains (see for eg. Kessel and Gruss, 1991). Therefore, the rib abnormalities were interpreted as being secondary to alterations in the lateral plate mesoderm.

Among Hoxd-12 transgenic mice with limb abnormalities, some animals (8 out of 26) displayed a severe phenotype that included transformations of digit I to a very long triphalangeal digit (digit III-like morphology) and frequent associated duplications of anterior digits, resulting in an appearance of partial mirror symmetry (Fig. 4A-C; Table 2). Many of these transgenics also displayed associated tibial hemimelia (shortening). Such phenotypes (mirror-symmetric duplications, transformations of anterior digits, associated hemimelia) were very reminiscent of several naturally occurring luxoid/luxate mouse mutants (Carter, 1951; Forsthoefel, 1962; Johnson, 1967) and also suggested the presence of ectopic anterior polarizing activity. In fact, ectopic Shh expression and ZPA activity have recently been demonstrated in several different luxoid mutants (Chan et al., 1995; Masuya et al., 1995, 1997).

We examined Shh expression in early embryos from line 2917 matings. In a subset of embryos, an ectopic focus of Shh expression was detectable by whole-mount in situ hybridization in either the anterior hindlimb or forelimb bud (Fig. 4E,F). This was never observed in parallel hybridizations with control embryos. In multiple hybridizations (20 litters total), about 12% of the expected homozygous embryos displayed an ectopic focus of Shh in anterior limb bud, compared to a 22% occurrence of mirror-symmetric digital patterns in the predicted number of line 2917 homozygotes. Thus, ectopic Shh expression was seen with about half the frequency of mirror-symmetric limb phenotypes. This imperfect correlation may be related to some variability in embryonic stages (which ranged from E9.5-E13) and/or lower sensitivity of the whole-mount in situ detection compared to the level of Shh expression necessary for bioactivity (abnormal polarization). The occurrence of mirror-symmetric limb changes in only a subset of phenotypically affected transgenics (8 out of 29 analyzed) could reflect variable transgene expression, with only the higher levels of Hoxd-12 misexpression resulting in a high enough level of Shh activation to alter limb polarization. Notably, ectopic foci of Shh expression were usually seen subjacent or near to the AER, suggesting involvement of AER signals together with Hoxd-12 misexpression.

The apparent correlation between mirror-symmetric limb phenotypes and ectopic Shh expression in a subset of transgenic embryos suggested that Hoxd-12 can activate Shh. We examined the Shh promoter to assess whether such activation of Shh might be a direct effect of Hoxd-12 misexpression.

Hoxd-12 transactivates through binding to either a typical TAAT motif or to a variant TTTAY motif (R. Hutson, S. Aguanno and S. Mackem, unpublished data); a variant TTAT core is also preferred by several other Abd B subtype homeobox genes (Benson et al., 1995; Ekker et al., 1994). 1 kbp of DNA upstream from the transcribed region of the mouse Shh gene was sequenced and found to contain several Hoxd-12 consensus binding sites, all within the first 700 bp, as well as a TATA motif in the expected location (Fig. 5). This DNA was ³²P-end-labeled, bound to recombinant Hoxd-12/GST fusion protein and digested with T7 exonuclease (T7 exo) to map Hoxd-12-binding sites. Four out of eight consensus matches identified by sequence comparison were adjacent to T7 exo stop sites generated by Hoxd-12 binding (Fig. 6A-C), while two were not, and two could not be evaluated due to overlap with the position of T7 exo limit-digestion product of naked DNA. Of the total of six exonuclease stops detected, only one was not closely positioned near a good Hoxd-12 consensus site match in the promoter region analyzed.

The 700 bp Shh promoter region containing Hoxd-12binding sites was ligated to CAT coding sequences to assay expression driven by the Shh promoter in cell culture. This reporter was transfected into primary cultured early limb bud cells from 72 hour chick embryos. Cotransfection with a Hoxd-12 expression vector reproducibly stimulated expression of the Shh-CAT reporter an average of 2.5-fold over baseline, suggesting that Hoxd-12 can transactivate expression of the Shh promoter (Figs 6D, 7). High baseline expression was not simply due to endogenous Hoxd-12 in the limb cells, as this was observed even when cells from only anterior limb bud halves were used (not shown), and may reflect the presence of elements permitting inappropriate basal Shh promoter activity from the fragment used, because other presumptive negative regulatory elements are missing. To confirm that the activation by Hoxd-12 was mediated by direct binding to the Shh promoter, the full-length Hoxd-12 protein fused to the potent VP16 activation domain was tested and found to strongly stimulate expression of the Shh-CAT reporter (Fig. 7).

To obtain independent confirmation of the ability of Hoxd-12 to regulate the natural Shh promoter, we assessed whether infection with a Hoxd-12-expressing retrovirus could activate the resident Shh gene in cultured chick limb bud mesenchymal cells. Since Hoxd-11 misexpression in chick results in phenotypes somewhat similar to the Hoxd-12 transgene (leg triphalangeal digit I, shortened tibia, wing digit duplication; see Morgan et al., 1992), it was of interest to determine whether this Hoxd member might also be able to regulate Shh expression. Cultured limb cells were infected with either Hoxd11- or Hoxd-12-expressing retrovirus and, after 48 hours, were analyzed for Shh transcript levels. As shown if Fig. 8, the introduction of either Hoxd-11 or Hoxd-12 in these cells stimulates expression of the endogenous Shh gene compared to the retroviral expression vector alone. Interestingly, this induction was dependent upon inclusion of FGF in the culture, again suggesting that AER signals may cooperate with Hoxd genes to induce Shh expression.

Normal Hoxd-12 expression begins in the posterior half of the limb bud mesenchyme and later becomes more distally restricted and also extends more anteriorly within the distal presumptive digit region, but never encompasses digit I (Yokouchi et al., 1991; Kondo et al., 1996, Nelson et al., 1996). Hoxd-12 misexpression phenotypes resemble ‘posterior’ transformations in some respects. In the distal autopod (hand/foot), where the transgene is expressed in both the forelimb and hindlimb, it is the anterior part of the autopod that is affected; anterior digits (eg. digit I) are converted to more posterior digits and/or become duplicated. In the hindlimb, where the Hoxd-12 transgene is also expressed proximally, the anterior-most element in the distal row of tarsals displays ‘posterior’ transformations as well. However, the anterior long bone (tibia) is shortened. Likewise, the phylogenetically ‘anteriormost’ pubic bone in the pelvis is reduced (discussed further below). At the same time, other proximal elements (femur, ilium) are unaffected.

There are several salient features of the phenotype. (1) The effects of Hoxd-12 on the skeletal pattern are evident as soon as condensations become visible and affect the formation of condensations as well as their subsequent growth, as evidenced by digit duplications and triphalangeal digit I transformations. (2) The transgene exerts its effects exclusively in domains where endogenous Hoxd-12 is not normally expressed: the anteriormost part of the autopod (digit I region), the anterior zeugopod (tibia) and the anterior pelvis. This suggests that Hoxd-12 levels are already saturating in regions where it is normally expressed, as observed for other Hox genes in the limb (Morgan et al., 1992; Yokouchi et al., 1995; Goff and Tabin, 1997). (3) In domains where Hoxd-12 is not normally expressed, the transgene promotes, represses, or has no effect on the formation and growth of chondrogenic condensations, depending on the particular locale. In contrast to effects on the autopod, reductions occur in the tibia and pubic bone while other proximal elements that are also outside of the normal Hoxd-12 expression domain are unaltered (femur and ilium). Such results are not easily reconciled with ‘homeosis’ models.

The formation of skeletal elements from proximal to distal (hip to toe) proceeds by progressive branching and segmentation of chondrogenic condensations to produce the more distal elements (eg. the femur branches distally to produce tibia and fibula). Comparisons of the branching pattern in various tetrapods has generated a model in which the autopod forms by an anterior bending of the main (metapterygial-like) limb axis: the distal row of carpal/tarsals and the digits all arise as successive postaxial branching events from the continuation of the main limb axis along a curving ‘digital arch’ (Shubin and Alberch, 1986). In this view, the digits (including digit I) and the distal row of tarsals are in fact all ‘postaxial’ structures whereas the tibia or the radius are true preaxial branches (see Fig. 9). The effects of Hoxd-12 misexpression in the limb are compatible with a model in which Hoxd-12 promotes formation of postaxial condensations branching from the main limb axis (previously proposed for Hoxd genes; see Duboule, 1994), while selectively inhibiting formation or growth of preaxial condensations.

The pelvis arises from the lateral plate mesoderm (Chevallier, 1977) and has traditionally been considered a part of the hindlimb. Its development is less well understood, partly because it is specified prior to formation of a visible limb bud swelling and the early condensations are ill-defined (see for eg. Rogulska, 1965). Work with a chick mutant indicates that the pelvis can form in the absence of apical ridge and polarizing signals (Prahlad et al., 1979; Ros et al., 1996; Noramly et al., 1996; Grieshammer et al., 1996). In mouse and human, the pelvis and femur appear to arise by segmentation from a single condensation (Forsthoefel, 1963; Rooker, 1979), making the pelvic anlage part of the branching/segmenting limb axis. The pubic bone of the pelvis is phylogenetically its anteriormost component (rotations in mammals distort this relationship; Hinchliffe and Johnson, 1980), suggesting the pubis may represent a ‘preaxial’ branch from a pelvic condensation. Since Hoxd-12 misexpression often results in pubic reduction in the affected pelvis, this phenotype may also be compatible with a model in which Hoxd-12 exerts inhibitory effects on preaxial condensations (Fig. 9).

Misexpression of Hoxd-13 or Hoxa-13 in the chick causes reductions in proximal long bones, where these genes are not normally expressed (Yokouchi et al., 1995; Goff and Tabin, 1997). Hoxd-11 misexpression in the chick produces a somewhat similar phenotype to that of transgenic Hoxd-12, including conversions of digit I to a triphalangeal digit and marked tibial reductions (Morgan et al., 1992; Goff and Tabin, 1997). Goff and Tabin (1977) have proposed that Hoxd-11 promotes growth of distal elements (digits) and represses growth of more proximal elements (long bones), since mild reductions in the fibula and femur were also observed with Hoxd-11 misexpression. However, transgenic Hoxd-12 selectively retards tibial development and concomitant bowing of the fibula suggests that any mild fibular reduction results from the tibial remnant acting as a mechanical tether to impede normal growth. The relatively normal femur in Hoxd-12 transgenics also indicates that proximal skeletal elements are not affected uniformly. Mild shortening of both forelimb zeu-gopodal long bones (radius and ulna) in Hoxd-12/Hoxa-11 null mice (Davis and Capecchi, 1996) suggests that Hoxd-12 may play some positive role in development of both zeugopodal long bones (tibia and fibula in hindlimb). In these loss-offunction mutants, the preaxial branch may be secondarily affected due to a primary reduction in the postaxial branch/main limb axis from which it bifurcates.

How opposing effects of Hoxd-12 on different condensations may be mediated is unknown. When inappropriately expressed, Hoxd-12 may interfere with the function of other Hox genes proximally, as proposed for Hoxd-13 (van der Hoeven et al., 1996; Goff and Tabin, 1997). However with Hoxd-12, such dominant-negative effects would be restricted to preaxial condensations (pubis, tibia, as compared to ilium, femur). Furthermore, it is possible that any, or perhaps all, of the stimulatory and inhibitory effects of Hoxd-12 on chondrogenic condensations may result indirectly from feedback induction of Shh, since very similar selective changes in the digits, tibia and pubis occur in several luxoid mouse mutants that misregulate Shh (see below).

In Hoxd-12 transgenic mice, reduction in the number of sternal ribs from seven to six and fusions of anterior sternal ribs could be considered posterior transformations, usually thought to result from an altered Hox code in the somitic mesoderm from which the ribs arise (see, for example, references in Krumlauf, 1994). Since the Hoxd-12 transgene is expressed solely in the lateral plate mesoderm, in this case, the rib changes must be secondary, perhaps related to altered signals from the lateral plate or to associated sternal dysgenesis. The sternum develops from paired bands in the dorsal lateral plate mesoderm that condense and move ventrally to meet in the midline where they fuse and segment in association with the ribs (Chen, 1952a,b; Chevallier, 1977). While the sternal phenotype seen in Hoxd12 transgenics is often severest posteriorly (caudally), the formation and/or movement of the entire sternal band appears to be affected, ranging from split sternebrae to complete sternal agenesis. Primary effects of Hoxd-12 on formation and growth of sternal chondrogenic condensations may produce a phenotype resembling homeosis, as proposed for the generation of apparent homeotic transformations in the axial skeleton due to altered Hox gene expression (see Duboule, 1995). The sternal and rib changes in Hoxd-12 transgenics are also very similar to those seen in the Xt mutant, and may be mechanistically related (discussed below).

Mirror-symmetric limb phenotypes seen in some Hoxd-12 transgenic mice correlate with induction of ectopic Shh in the anterior limb bud. In vitro binding and activation of the Shh promoter, and induction of endogenous cellular Shh RNA by exogenous Hoxd-12 all suggest that Hoxd-12 may directly regulate Shh. Hoxd genes are thought to be downstream of polarizing signals (see Izpisua-Belmonte and Duboule, 1992) and misexpression of Shh induces de novo expression of Hoxd genes in the limb bud (Riddle et al., 1993). Thus, activation of Shh expression by Hoxd-12 may represent the return half of a positive feedback loop; such loops are often used to amplify signals, particularly in the limb (reviewed by Cohn and Tickle, 1996). In chick, work with a limbless mutant and with limbs made of reaggregated anterior mesenchyme indicate that Hoxd genes can be expressed in the absence of a Shh signal and are posteriorly polarized in the early limb bud in the absence of Shh (Grieshammer et al., 1996; Hardy et al., 1995; Noramly et al., 1996; Ros et al., 1994, 1996). This raises the alternative possibility that some other earlier asymmetric signal initially induces posterior 5′Hoxd gene expression (e.g. Hoxb-8; Charite et al., 1994) and that certain 5′ Hoxd genes first activate Shh expression in the limb bud. Shh might then in turn induce the transition to a late ‘de novo’ distal domain of Hoxd expression in the limb bud (discussed by Duboule, 1994; Nelson et al., 1996). In any case, the ultimate outcome is similar; Shh and Hoxd-12 activate each other, resulting in a positive feedback loop.

Hoxd-13 may not participate in this feedback loop, since misexpression phenotypes suggest no altered polarization (Goff and Tabin, 1997). Hoxd-11 misexpression in chick has produced a phenotype somewhat similar to transgenic Hoxd12 (Morgan et al., 1992), but molecular evidence of ectopic polarizing signals was not observed, perhaps due to lower or non-uniform expression levels. Like Hoxd-12, Hoxd-11 upregulates Shh expression in retrovirally infected limb cells in culture, suggesting that it too may participate in positive feedback regulation of Shh. In vivo feedback regulation by Hoxd-12 (and Hoxd-11) may also require coincident FGF signals from the AER. Normal Shh expression occurs in a posterodistal domain subjacent to the AER that is more restricted than the expression domains of Hoxd-11 and Hoxd-12 (Riddle et al., 1993; Nelson et al., 1996). Similarly, the ectopic Shh domain seen in the anterior limb bud of Hoxd-12 transgenic embryos was often closely associated with the AER. A corequirement of FGF for the induction of cellular Shh by retrovirally expressed Hoxd-11 or Hoxd-12 would also support this view.

Feedback regulation also complicates interpretation of Hoxd-12 transgene effects. Presumably, ectopic Shh induces expression of other Hoxd genes as well as other targets. Some, or all of the phenotypic changes seen could reflect indirect misregulation of these other ‘downstream’ genes. Identification of direct targets of Hoxd-12, as well as of Shh action, may help resolve this issue.

The Hoxd-12 transgenic phenotype in lateral plate derivatives strikingly resembles certain luxoid/luxate mouse mutants (Carter, 1951; Forsthoefel, 1962; Hinchliffe and Johnson, 1980; Johnson, 1967; Masuya et al., 1995; Mo et al., 1997). These mutants all have triphalangeal digit I and varying degrees of anterior digit duplications with hemimelia (selective shortening), usually of the anterior zeugopodal long bone (tibia or radius). Strong’s luxoid (lst), Carter’s luxate (lx) and extra toes (Xt), also have selective reductions in the pubic bone of the pelvis; particularly loss of the superior pubic ramus and open symphysis (Carter, 1951; Forsthoefel, 1962; Johnson, 1967). Additionally, Xt has very similar sternal abnormalities, ranging from mildly split sternebrae to agenesis with open rib cage and reductions to six sternal ribs (Johnson, 1967; Mo et al., 1997).

Similarities between luxoid mutants and Hoxd-12 transgenics also exist at a molecular level. Several, including lst, lx and Xt, express ectopic Shh in the anterior limb bud (Chan et al., 1995; Masuya et al., 1995, 1997), suggesting that the constellation of hemimelia and ‘preaxial’ (anterior) digital polydactyly may generally indicate aberrant Shh signaling. Xt results from a loss-of-function mutation in the Gli3 zinc finger gene (Schimmang et al., 1992; Hui and Joyner, 1993), thought to function as a feedback repressor of Shh (Buscher et al., 1997; Marigo et al., 1996; Masuya et al., 1997; Mo et al., 1997). Xt mutants are haplo-insufficient, and heterozygotes and homozygotes resemble milder and more severe Hoxd-12 transgenic phenotypes, respectively (Johnson, 1967; Mo et al., 1997). The strikingly similar sternal, rib and pelvic changes seen in Xt and in Hoxd-12 transgenics suggest that these phenotypes in both may also result from altered Shh expression. If so, then detection of ectopic Shh in only a subset of Hoxd-12 transgenics must reflect a lower detection sensitivity than bioeffect threshold; such a difference has also been seen between ectopic Shh detection and phenotype in Xt (Masuya et al., 1995).

Given similar DNA binding by different Abd B type Hox genes in vitro (Benson et al., 1995; Ekker et al., 1994), might Hoxd12 be mimicking the effect of some other posteriorly expressed Hox gene in vivo? Probably not, for several reasons. (1) In vivo DNA recognition must be considerably more specific than in vitro. In gain-of-function analyses of several Abd B type Hox genes in chick, including Hoxd-10, Hoxd-11, Hoxd-13 and Hoxa-13 (Morgan et al., 1992; Goff and Tabin, 1997; Yokouchi et al., 1995), only Hoxd-13 and Hoxa-13 have similar phenotypes. Of this group, only Hoxd-11 phenotypically resembles Hoxd-12 transgenics, which may reflect bona fide functional overlap in vivo. (2) There is ample evidence supporting the view that, particularly in the limb, Hox genes are present at saturating levels in their normal expression domains, since over-expression at these sites produces no phenotype (Morgan et al., 1992; Goff and Tabin, 1997; Yokouchi et al., 1995; this study). However, if different 5′ Hox genes can all substitute for one another when overexpressed (by binding each others targets), then misexpression of any Abd B type Hox gene should always produce a similar phenotype, regardless of the normal expression domain of that particular gene and saturation should not occur. (3) While Hoxb-8 misexpression in transgenic mice induces Shh and produces some similar phenotypes to Hoxd12 (Charite et al., 1994), this gene belongs to the Abd A type, whose DNA-binding properties, even in vitro, are different from Abd B type Hox genes (see Ekker et al., 1994 and references therein). The similarity more likely reflects the in vivo complexity of Shh regulation.

Consequently, the Hoxd-12 misexpression phenotype most likely reflects in vivo roles for this gene in patterning chondrogenic elements and in amplifying polarizing signals during limb bud outgrowth via a positive feedback loop with Shh. A number of different genes have been implicated in regulating Shh expression in the limb (including some 5′ Hoxd genes, Hoxb-8, Gli3 and other luxoid mutant genes, as well as FGF and Wnt members), and may all be required in the complex regulatory circuitry modulating Shh expression in order to achieve precise and dynamic temporospatial regulation of this important signaling molecule.

We thank D. Levens and C. Tabin for comments and discussions; M. Federspiel and S. Hughes for advice on propagating retroviruses; C. Baker, D. Duboule, T. Kristie, A. McMahon, R. Riddle, and C. Tabin for probes and constructs; Genetics Institute, Inc. for recombinant FGF-4 protein; and W. Randolph and G. Best for expertise in photography and computer imaging. V. K. is a visiting fellow on leave from the University of Zagreb School of Medicine.