Scapula development is governed by genetic interactions of Pbx1 with its family members and with Emx2 via their cooperative control of Alx1

The development of the shoulder blade (scapula) is poorly understood. In the chick, its proximal components, i.e. the blade and spine, derive from dermomyotomal mesenchyme, whereas its distal components, i.e. the glenoid cavity and coracoid, as well as the acromion, derive partly from the somatopleuric compartment of the lateral plate mesoderm (LPM) (Huang et al., 2000; Wang et al., 2005) (see Fig. 2). In the mouse, the bony structures of the scapula have dual neural crest-mesoderm origin, as shown by genetic lineage labeling (Matsuoka et al., 2005).

Scapula development is modestly affected by the genetic pathways that pattern the limb (e.g. Prahlad et al., 1979; Ros et al., 1996), and there is only rudimentary knowledge of the genes that act upstream of blade and head development (Huang et al., 2006). By contrast, much information has been gathered on genes that finely pattern these structures. For example, Pax1/Hoxa5 and Hoxc6 are involved in acromion and head development, respectively (Aubin et al., 1998; Timmons et al., 1994). Additionally, two distinct pathways have been suggested in blade patterning (Kuijper et al., 2005): one regulated by Emx2, the loss-of-function of which results in blade absence in mice (Pellegrini et al., 2001), and the other governed by genetic interactions among aristaless-related (Alx1, Alx3, Alx4), T-box (Tbx15) and Gli (Gli3) genes. It remains unknown whether these two blade-specific pathways are separate or integrated, whether Emx2 controls one or both, and which genes act upstream of Emx2 in blade and head morphogenesis.

Recent studies suggest that Pbx TALE homeoproteins control scapula development (Capellini et al., 2006; Selleri et al., 2001). Indeed, Pbx1-null (Pbx1^–/–) embryos exhibit hypoplastic blades and scapular head-humeral fusions (Selleri et al., 2001), whereas compound Pbx1^–/–;Pbx2^+/– mutants present exacerbations of these phenotypes (Capellini et al., 2006; Capellini et al., 2008). Conversely, the single loss of Pbx2 or Pbx3 does not cause scapula mutant phenotypes (Rhee et al., 2004; Selleri et al., 2004). Scapula and proximal limb have thus been proposed to be `Pbx dependent', consistent with Pbx roles as Hox co-factors (reviewed by Moens and Selleri, 2006). However, loss of Hox genes, including entire Hox clusters, from pre-scapular tissues results, at best, in negligible scapula alterations, despite co-expression of Pbx and Hox genes in these domains (Fromental-Ramain et al., 1996; Aubin, 1997). Therefore, it currently appears unlikely that Pbx proteins function as Hox co-factors in scapula formation, suggesting instead cooperation with other factors.

Emx2 shares structural characteristics with Hox proteins, including: (1) a Gln (Q) in position 50 of the homeodomain, which is predicted to confer DNA-binding properties similar to those of Hox proteins; and (2) a YPWL sequence N-terminal to the homeodomain that resembles part of the Hox IYPWMK motif [`hexapeptide' (Bürglin, 1994)], a hallmark for heterodimerization with Pbx (reviewed by Moens and Selleri, 2006). It has never been addressed, though, whether Pbx factors and Emx2 interact genetically and/or physically in scapula development.

Here, we demonstrate that Pbx genes share overlapping functions in scapula development, as Pbx1;Pbx2 and Pbx1;Pbx3 compound mutants display more severe and novel phenotypes compared with those of Pbx1^–/– embryos. We also show that the expression of genes crucial for scapular blade patterning or acromion and head development is altered in these mutants, placing Pbx genes upstream of pathways involved in scapula ontogeny. Additionally, we demonstrate that Pbx1 genetically interacts with Emx2 in blade and spine formation, as compound Pbx1;Emx2 mutants display novel defects. Moreover, we establish that Pbx1 and Emx2 heterodimerize on a specific DNA sequence, providing a molecular basis for their genetic interaction. Pbx1 and Emx2 proteins also bind in vivo to a conserved sequence upstream of the Alx1 (formerly Cart1) gene, and cooperatively activate transcription via this potential regulatory element. Our results thus delineate novel genetic networks in shoulder girdle development.

Intercrosses between Pbx1^+/– (Selleri et al., 2001), Pbx2^+/– (Selleri et al., 2004), Pbx3^+/– (Rhee et al., 2004) and Emx2^+/– (Pellegrini et al., 2001) mice were performed to obtain Pbx1^+/–;Pbx2^+/–, Pbx1^+/–;Pbx3^+/–, Pbx2^+/–;Pbx3^+/– and Pbx1^+/–;Emx2^+/– mutants. On a C57BL/6 background, the double-heterozygous numbers obtained were below the expected Mendelian ratios. To increase numbers, C57BL/6 double-heterozygous males for each mutant line were crossed to Black-Swiss [NIH-BL(S)]. Next, double-heterozygous C57BL/6 females and mixed double-heterozygous C57BL/6-Black-Swiss males were intercrossed and their progeny analyzed for skeletal phenotypes. On mixed genetic backgrounds, marked ameliorations of C57BL/6 scapular phenotypes were observed, which led us to use C57BL/6 to Black-Swiss intercrosses. To generate more complex Pbx compound genotypes, Pbx1^+/–;Pbx2^+/– and Pbx1^+/–;Pbx3^+/– mutants were intercrossed to obtain Pbx1^+/–;Pbx2^+/–;Pbx3^+/– mice; their in utero lethality prevented further intercrossing.

Skeletons of E13.5-14.5 embryos were stained with Alcian Blue/Alizarin Red S (Depew et al., 1999; McLeod, 1980). Whole-mount in situ hybridizations were performed on somite-matched embryos using digoxygenin-labeled RNA probes (Di Giacomo et al., 2006). Alx1 (Kuijper et al., 2005), Alx4 (Beverdam and Meijlink, 2001), Emx2 (Pellegrini et al., 2001), En1 (Davidson et al., 1988), Gli3 (Beverdam and Meijlink, 2001), Hoxc6 (Sharpe et al., 1988), Pax1 (Chalepakis et al., 1991), Pbx1 (Brendolan et al., 2005), Pbx2 (Selleri et al., 2004), Pbx3 (Di Giacomo et al., 2006), Sox9 (Wright et al., 1995) and Tbx15 (Kuijper et al., 2005) probes were used. Section in situ hybridization was conducted as described (Di Giacomo et al., 2006). Cell proliferation and apoptosis assays were carried out as described (Capellini et al., 2008); for details, see Fig. S1 in the supplementary material. Skeletal preparations and whole-mount in situ hybridization embryos were processed for OPT according to published protocols (Sharpe, 2003; Sharpe et al., 2002).

As previously described (Salsi et al., 2008), chromatin was extracted from mouse proximal forelimb and flank dissected from wild-type E11.5 embryos. Cross-linked chromatin was immunoprecipitated with 5 μg anti-Pbx antiserum (C-20, Santa Cruz) or a specific anti-Emx2 monoclonal antibody (see below). Semi-quantitative PCR analysis was performed on three independent ChIPs for each genomic region analyzed, and primers were designed accordingly (Salsi and Zappavigna, 2006) (see Table S1 in the supplementary material).

The pSG5Pbx1 and pETPbx1 expression constructs for the Pbx1a protein have been described (Di Rocco et al., 1997). The pSG5Emx2 expression construct was generated by cloning EcoRI/BglII-cut PCR-amplified Emx2 coding sequence into EcoRI/BamHI-cut pSG5 (Stratagene) vector. The pRSETEmx2 expression construct was generated by cloning the Emx2 coding sequence as an EcoRI fragment into the EcoRI site of the pRSETB vector. The pML(EP)₆ reporter construct was obtained by cloning a tandem hexamer of the EP oligonucleotide 5′-GATCCGTCGACGGATCATTAAAGCCCTCGAGA-3′ into the BglII site of the pML luciferase reporter (Di Rocco et al., 1997). The pMLAlx1-EPBS(900) reporter was generated by cloning a 974 bp fragment containing the PCR-amplified Alx1 5′-7 genomic sequence [–4670 to –3699, relative to the transcription start site (TSS)] into the SmaI site of pML. The pMLAlx1-EPBS reporter was generated by cloning a 130 bp PCR-amplified fragment from the Alx1 5′-7 region containing Sites A, B and C of the EPBS sequence (–4270 to –4139) into the NheI/BglII sites of pML. The pMLAlx1-EPBS(900)A+Bmut and pMLAlx1-EPBS mutated reporters (Amut, Bmut or Cmut) were obtained by mutating Site A, B, A+B or C within the EPBS element by PCR-mediated mutagenesis and cloning the fragments into the NheI/BglII sites of pML. The mutations were identical to those of the electrophoretic mobility shift assay (EMSA) oligonucleotide probes (see below). All PCR-amplified fragments were verified by sequencing.

His-tagged Emx2 and Pbx1 proteins were produced in E. coli using pRSETEmx2 and pETPbx1 plasmids and purified using Ni-NTA Agarose (Qiagen). For binding site selection, purified Emx2 and Pbx1 proteins were incubated with ³²P-labeled SelEP oligonucleotide pool (5′-GGCGAGATCTCTCGAGGGNNNNNATGATCCGTCGACGGATCCGCGG-3′) in binding buffer (Chang et al., 1995) for 30 minutes at 0°C and electrophoresed on a 6% polyacrylamide gel in 0.5× TBE. The retarded band corresponding to the Emx2-Pbx1 heterodimer was excised and eluted in 0.5 M ammonium acetate, 10 mM MgCl₂, 1 mM EDTA, 0.1% SDS. Aliquots of the eluted band were PCR amplified for 30 cycles (95°C 1 minute, 55°C 30 seconds, 72°C 1 minute) with primers complementary to the flanking arms of SelPE, and the amplified DNA was used in further binding reactions. After six selection rounds, amplified DNA was cloned, sequenced and consensus sequences were aligned using CLUSTAL W (Thompson et al., 1994) (http://workbench.sdsc.edu).

EMSAs were performed (Di Rocco et al., 2001) using purified, bacterially produced Emx2 and Pbx1 proteins, or whole-cell extracts (10-20 μg each) from COS cells exogenously expressing both genes. To identify protein-DNA complexes, 200 ng of anti-Pbx antibody (C20, Santa Cruz) or 0.1 μl of an anti-Emx2 monoclonal antibody were added to binding reactions, which were separated by 6% PAGE in 0.5× TBE, dried and exposed to film. The anti-Emx2 monoclonal antibody was generated using the N-terminal portion of Emx2 and tested for its lack of cross-reaction with Pbx1 by immunoblotting. Oligonucleotides used in EMSAs were (mutated nucleotides are underlined):

• Alx1-EPBS, 5′-TATAGCTTTGATGTAAGTAGAAGTATCTTTCATGTCCAAAATTAAAATTACATT-3′;

• EPBS-Amut, 5′-TATAGCGGGGCGGGCAGTAGAAGTATCTTTCATGTCCAAAATTAAAATTACATT-3′;

• EPBS-Bmut, 5′-TATAGCTTTGATGTAAGTAGAAGTATCTGGACGGTCCAAAATTAAAATTACATT-3′; and

• EPBS-Cmut, 5′-TATAGCTTTGATGTAAGTAGAAGTATCTTTCATGTCCAAACGGCAACGGCCATT-3′.

P19 mouse embryonal carcinoma cells were cultured in MEMα with 10% foetal calf serum. In typical transfections, performed by CaPO₄ precipitation, reporter plasmid (5 μg), protein expression construct (2-4 μg), pCH110 β-galactosidase plasmid (0.5 μg) as an internal control, and pBluescript (variable quantities), were added to a total of 20 μg transfected DNA per 9-cm dish. Cells were harvested 48 hours post-transfection, and assayed for luciferase and β-galactosidase expression (Zappavigna et al., 1994).

At E10.5, Pbx1 and Pbx2, but not Pbx3, were expressed in the dermomyotome and somatopleure (see Fig. S1A in the supplementary material). However, from E11 to E12, all three Pbx genes were detected, although Pbx3 was at lower levels (see Fig. S1B in the supplementary material). At E10.5-11.5, all three Pbx genes were co-expressed in proximal forelimb mesenchyme (Fig. 1A,C), corroborating previous reports (Capellini et al., 2006). This pattern persisted until E11.0-11.5, when the expression of Pbx1 remained proximal, whereas Pbx3 was confined to proximal-anterior limb domains and Pbx2 to distal domains (see Capellini et al., 2006; Di Giacomo et al., 2006). Emx2 was co-expressed with Pbx genes at E10.5 (Fig. 1A,C) (Theil et al., 1999; Tian and Lev, 2002).

Given the overlapping expression of the Pbx genes in pre-scapular domains, compound Pbx mutants were generated to investigate the respective contributions of Pbx1-3. Scapula defects were evident in only three allelic combinations: Pbx1^–/–, Pbx1^–/–;Pbx2^+/– and Pbx1^–/–;Pbx3^+/– (Fig. 2A,C; see Fig. S2A in the supplementary material); Pbx1^–/–;Pbx2^–/– and Pbx1^–/–;Pbx3^–/– mutants died in utero before scapula morphogenesis (Capellini et al., 2006).

At E13.5, Pbx1^–/– embryos exhibited reduced blades and spines, obliterated glenoid cavities fused to humeral heads and expanded coracoids (Fig. 2A) (Selleri et al., 2001). In Pbx1^–/–;Pbx2^+/– mutants (Pbx1/2Mut), several Pbx1^–/– scapula phenotypes were exacerbated and novel defects became evident. Cartilage preparations at E12.5 revealed that scapular condensations were disorganized and reduced, indicating disruption prior to mesenchyme formation (Capellini et al., 2006). Analysis of Sox9 expression, a mesenchymal condensation marker, revealed significant reductions across the superior-to-inferior blade-forming domains (Fig. 3A). OPT (Sharpe, 2003; Sharpe et al., 2002) of the three-dimensional expanse of Sox9 expression revealed a decrease in mesenchyme across entire pre-scapular domains (Fig. 3C; see Fig. S2B and Movie 1 in the supplementary material). As a result, reduced cartilages were observed at E12.5/E13.5 in skeletal preparations and by OPT. Pbx1/2Mut scapulae (Fig. 2A,C; see Fig. S2A and Movie 2 in the supplementary material) were dysmorphic, their blades were markedly reduced compared with those of Pbx1^–/– and wild-type (WT) embryos, and their only intact aspects were near the head-neck junction. Furthermore, spines and acromia were absent. The remaining `scapular' mass was fused to the humeral head with an adjoining skeletal element, possibly a duplicated scapular/humeral head (Fig. 2A, inset, Fig. 2C; see Fig. S2A in the supplementary material). At E13.5, skeletal preparations and OPT revealed that Pbx1^–/–;Pbx3^+/– (Pbx1/3Mut) embryos also exhibited scapulae with proximal-to-distal blade reductions (Fig. 2A,C; see Fig. S2A in the supplementary material), although less severe than in Pbx1/2Mut embryos (Fig. 2A,C; see Fig. S2A and Movie 3 in the supplementary material). Additionally, Pbx1/3Mut scapulae showed blade indentations (Fig. 2C; see Fig. S2A in the supplementary material) and foramina (Fig. 2A), lacked well-developed spines and acromia, and their heads remained fused to humeri and were potentially duplicated (Fig. 2A,C; see Fig. S2A in the supplementary material). Sox9 expression was reduced in Pbx1/3Mut scapular domains (Fig. 3A; see Fig. S2B in the supplementary material), although OPT revealed that this reduction was not as drastic as in Pbx1/2Mut embryos (Fig. 3C; see Fig. S2B and Movie 4 in the supplementary material).

Scapular domains of Pbx2;Pbx3 mutants, particularly Pbx2^–/–;Pbx3^+/– and Pbx2^+/–;Pbx3^–/– embryos, appeared similar to those of WT (not shown), whereas Pbx2^–/–;Pbx3^–/– mutants died before shoulder development could be assessed (i.e. before E10.5). Finally, Pbx1;Pbx2;Pbx3 compound mutants could not be obtained owing to the perinatal lethality of Pbx1^+/–;Pbx2^+/–;Pbx3^+/– embryos, which did not exhibit appendicular defects (not shown).

In light of the blade defects of compound Pbx mutants, we examined the expression of Emx2, the absence of which results in blade agenesis (Pellegrini et al., 2001). Whereas Emx2 was unchanged in E10.5 Pbx1^–/– embryos (not shown), it was severely and moderately reduced in Pbx1/2Mut and Pbx1/3Mut embryos, respectively (Fig. 1B). By contrast, Pbx1-3 expression remained grossly unperturbed in the proximal-superior forelimbs of Emx2^–/– embryos (see Fig. S1C in the supplementary material).

Given that (1) Pbx and Emx2 genes are co-expressed in blade-patterning domains, (2) mice deficient for either of these genes exhibit blade phenotypes, and (3) both Pbx and Emx2 proteins possess structural features permitting their heterodimerization, potential genetic interactions of Pbx1 with Emx2 were assessed. At E14.5, Pbx1^+/+;Emx2^–/– embryos displayed blade loss, as previously reported (Fig. 2B) (Pellegrini et al., 2001), whereas Pbx1^–/–;Emx2^+/+ embryos exhibited blade alterations and head fusions (Fig. 2A) (Selleri et al., 2001). Genetic interaction was evident in Pbx1^+/–;Emx2^+/– embryos, which showed minor proximal blade reductions that were absent in single Pbx1^+/– or Emx2^+/– progeny (not shown), and in Pbx1^–/–;Emx2^+/– mutants (Pbx1/Emx2Mut), which exhibited centrally bifurcated blades with superior and inferior rami resembling individual bones (Fig. 2B). OPT uniquely revealed rounded, shaft-like rami, rather than flattened, blade-like structures (Fig. 2C; see Fig. S2A and Movie 5 in the supplementary material). These phenotypes were not observed in Pbx1^–/– or Emx2^+/– embryos (Fig. 2B,C). Additionally, Pbx1/Emx2Mut limbs displayed scapular-humeral head fusions, as in Pbx1^–/– mutants (Fig. 2B,C; see Fig. S2A in the supplementary material) (Selleri et al., 2001). Analysis of Sox9 expression in Pbx1/Emx2Mut embryos at E12.5 revealed reduced mesenchymal condensations, and OPT revealed this mesenchyme to be markedly indented along its vertebral border, foreshadowing later scapular bifurcation (Fig. 3C; see Fig. S2B and Movie 6 in the supplementary material). Finally, Pbx1^–/–;Emx2^–/– mutants displayed blade agenesis, as in Emx2^–/– embryos, and scapular-humeral head fusions, as in Pbx1^–/– embryos (Fig. 2B). Given the additive phenotype of Pbx1^–/–;Emx2^–/– mutants, only Pbx1/Emx2Mut mice were studied for scapula marker gene expression.

To investigate Pbx and Emx2 roles in patterning distal scapular structures, the expression of genes crucial for head and acromial development was examined in compound mutants. Pax1, a regulator of acromion formation (Aubin et al., 2002; Timmons et al., 1994; Wilm et al., 1998), was markedly reduced proximally (Fig. 4A, left) but was maintained within its dorsoventral domain (Fig. 4B, left) in Pbx1/2Mut and Pbx1/3Mut embryos, whereas it was only slightly decreased in Pbx1^–/– mutants (Fig. 4A, left). Severe reductions were not present in Emx2^–/– or Pbx1/Emx2Mut embryos (Fig. 4A, left). Although Hoxc6^–/– mutants lack scapular head malformations (Garcia-Gasca and Spyropoulos, 2000), misexpression studies implicate this gene in its specification (Oliver et al., 1990). In Pbx1^–/–, Pbx1/2Mut and Pbx1/3Mut embryos, Hoxc6 was posteriorly expanded in limbs, unlike in controls (Fig. 4A, right) and was dorsoventrally expanded in compound Pbx mutants (Fig. 4B, right), but unaltered in Emx2^–/– or Pbx1/Emx2Mut embryos (Fig. 4A, right).

Only the proximal forelimb expression of regulators of blade patterning was considered crucial (Fig. 5; see Figs S3 and S4 in the supplementary material), as distal domains typically correlate with limb development (Huang et al., 2006). The expression of Alx1, a gene that patterns the superior blade (Kuijper et al., 2005), was reduced in Pbx1^–/– (Fig. 5A, left) and Emx2^–/– (see Fig. S3 in the supplementary material) proximal forelimbs, whereas it was absent in E10.5-11.5 Pbx1/2Mut, Pbx1/3Mut and Pbx1/Emx2Mut proximal forelimbs (Fig. 5A, left). At E10.5, the expression of Alx4, which also patterns the superior blade, was only slightly reduced in Pbx1^–/– and Pbx1/Emx2Mut proximal forelimbs, but markedly reduced and spatially altered in Pbx1/2Mut and Pbx1/3Mut forelimbs (Fig. 5A, right). However, this perturbation was not as extreme at E11.5 in all mutants (see Fig. S4 in the supplementary material), including Emx2^–/– mutants (see Fig. S3 in the supplementary material). Expression of Tbx15, a gene that patterns the central blade, was only reduced in Pbx1/2Mut proximal forelimbs at E10.5 (Fig. 5B, left), whereas at E11.5 it was expanded in proximal forelimbs of Pbx1^–/– embryos and all other mutants lacking Pbx1 (see Fig. S4 in the supplementary material) but was absent in Emx2^–/– mutants (see Fig. S3 in the supplementary material). Expression of Gli3, which patterns the inferior blade, was nearly absent and reduced in Pbx1/2Mut and Pbx1/3Mut proximal forelimbs, respectively, from E10.5 (Fig. 5B, right) through E11.5, when only Pbx1/Emx2Mut (see Fig. S4 in the supplementary material) and Emx2^–/– (see Fig. S3 in the supplementary material) embryos retained normal Gli3 expression. The expression of Gli3 and Alx4, along with engrailed 1 (En1), which demarcates the somitic dermomyotome (Davidson et al., 1988), was unaltered in all of these mutants (not shown). These results are summarized in Fig. 8A.

Finally, proliferation and apoptosis assays were carried out in Pbx1/2Mut and Pbx1/Emx2Mut embryos (see Fig. S5 in the supplementary material; data not shown). No marked differences in cell proliferation, as assessed by in vivo BrdU incorporation, and in apoptosis, as revealed by anti-caspase 3 immunohistochemistry, were identified in pre-scapular domains of Pbx1/2Mut versus WT embryos from E9.5 to E11.5 (see Fig. S5A in the supplementary material; data not shown). Similarly, there was no conspicuous increase in apoptosis in Pbx1/Emx2Mut embryos (data not shown). By contrast, Pbx1/Emx2Mut embryos displayed reductions in cell proliferation compared with WT (see Fig. S5B in the supplementary material), and more marked reductions were observed in Emx2^–/– embryos, which lack the blade altogether (Pellegrini et al., 2001).

Given the Emx2 and Pbx1 genetic interaction, we determined the as yet unexplored optimal DNA-binding consensus sequence for a putative Emx2-Pbx1 heterodimer. An oligonucleotide containing random nucleotides in the five positions 5′ to the Pbx1 TGAT core consensus binding sequence (Lu et al., 1995; Van Dijk et al., 1993), along with purified Emx2 and Pbx1 proteins, were used in site-selection experiments (see Materials and methods). Emx2 and Pbx1 proved to bind DNA as a heterodimer, preferentially to sites matching the consensus sequence 5′-CTTTAATGAT-3′ (hereafter, the `EP' binding site) (not shown). To confirm that Emx2 and Pbx1 cooperatively bind EP, EMSAs were performed using whole-cell extracts from COS cells exogenously expressing Emx2 or Pbx1, alone or in combination. Extracts containing both Emx2 and Pbx1 showed a retarded band (Fig. 6A, lane 8) that was supershifted or abolished by the addition of an anti-Emx2 or anti-Pbx1 antibody, respectively (Fig. 6A, lanes 9 and 10). Whereas Pbx1 alone did not bind the EP site (lanes 5 and 6), Emx2 alone formed a faster-migrating shifted complex (lanes 3 and 4).

To test whether Emx2 and Pbx1 activate transcription via the EP site, a luciferase reporter construct containing six EP copies upstream of the adenovirus major late (AdML) minimal promoter [pML(EP)₆] (Fig. 6B) was transiently transfected into P19 embryonal carcinoma cells with increasing amounts of Emx2 and/or Pbx1 expression constructs. Whereas Emx2 or Pbx1 alone could not activate transcription via EP, their co-expression led to a significant stimulation above reporter basal activity (Fig. 6B), indicating that they physically interact to form DNA-bound heterodimers on specific DNA sequences and cooperatively activate transcription.

As Emx2 and Pbx1 act upstream of Alx1 (Fig. 5A, left), we examined whether Alx1 transcription is directly regulated by Emx2 and/or Pbx1. We searched for EP consensus sequences within a region spanning from ∼10 kb 5′ to the putative Alx1 transcription start site (TSS) to ∼20 kb 3′ to its polyadenylation signal (PAS), using MatInspector (Cartharius et al., 2005; Quandt et al., 1995). A binding site matrix representing the EP consensus was generated using MatInd (Cartharius et al., 2005), based on our site-selection results (not shown). Because an evolutionarily conserved motif similar to the EP consensus was not identified within this region, we determined whether Emx2 and/or Pbx1 were bound in vivo to putative Alx1 regulatory sequences. ChIP was performed on sixteen genomic regions surrounding Alx1 exons, chosen because of their evolutionary conservation and gene proximity (Fig. 7A). Nine regions were 5′ to the Alx1 TSS, with the most upstream (5′-8) 9 kb away (Fig. 7A), and seven were 3′ to the Alx1 PAS, with the most downstream (3′-7) 16 kb away (Fig. 7A). ChIP on E11.5 WT mouse proximal forelimb and flank chromatin using anti-Pbx and anti-Emx2 antibodies revealed that only region 5′-7 (Fig. 7A,B), comprising –4472 to –4075 from the TSS, displayed significant enrichments for both Emx2- and Pbx1-immunoprecipitated chromatin (Fig. 7B). This 400 bp region and its conserved flanking sequences were analyzed by EMSA for binding by Emx2 and Pbx1 (not shown). Only a highly conserved 54 bp sequence overlapping the 3′ end of this region showed significant binding (Alx1-EPBS, Fig. 7C). The Alx1-EPBS sequence was bound weakly by Pbx1 alone and not by Emx2 alone (Fig. 7D, left, lanes 2 and 3). However, addition of both Emx2 and Pbx1 led to the formation of a retarded band (Fig. 7D, lane 4) that was supershifted or abolished by an anti-Emx2 or anti-Pbx1 antibody, respectively (lanes 5 and 6), demonstrating that Emx2 and Pbx1 cooperatively bind Alx1-EPBS. The Alx1-EPBS sequence contains a conserved TGAT motif (Site A) representing the Pbx1 binding consensus, a conserved TCAT motif (Site B) found within the EP site (ATCATTAAAG, complementary strand), and two non-conserved ATTA motifs (Site C). We mutated (see Materials and methods) these three putative core binding motifs (EPBS-Amut, EPBS-Bmut and EPBS-Cmut, Fig. 7D) and tested them in EMSAs. Mutation of TGAT (Amut, lanes 7-10) or TCAT (Bmut, lanes 11-14) motifs strongly reduced Pbx1-Emx2 cooperative binding, whereas mutation of the ATTA motifs (Cmut, lanes 15-18) only weakly reduced the formation of the Pbx1-Emx2 retarded complex.

To test whether Pbx1 and/or Emx2 activated transcription via the Alx1-EPBS sequence, we generated luciferase reporter constructs by inserting a 974 bp fragment spanning region 5′-7 (Fig. 7A) including the Alx1-EPBS sequence [EPBS(900), Fig. 7E], or a 130 bp fragment that exclusively encompasses the Alx1-EPBS sequence (EPBS, Fig. 7E), upstream of the AdML promoter. These reporter constructs were transfected together with Emx2 and/or Pbx1 expression constructs into P19 cells. Whereas Emx2 and Pbx1 alone did not appreciably stimulate the EPBS(900), or the EPBS, reporter activity, their co-expression led to an increase in EPBS(900) (∼5.9-fold) and EPBS (∼5.4-fold) reporter activity (Fig. 7E). We then generated reporter constructs carrying the same mutations tested in EMSAs within the single Site A (EPBS-Amut), Site B (EPBS-Bmut) or Site C (EPBS-Cmut), and a compound mutation of Sites A and B [EPBS(900)A+Bmut] (Fig. 7E). Whereas the mutation of Site C did not cause any reduction in reporter transactivation by Emx2 and Pbx1, the mutation of Site A or Site B caused a decrease in reporter transactivation (Fig. 7E). A further decrease in Emx2-Pbx1-mediated transcriptional activation was caused by the compound mutation of Sites A and B (Fig. 7E). These results indicate that both Sites A and B are necessary for the assembly of a DNA-bound Emx2-Pbx1 complex and for full transcriptional activation by Emx2-Pbx1 via the Alx1-EPBS sequence, whereas Site C is dispensable.

In dermomyotome, in which Pbx3 is not expressed, the compound absence of Pbx1 and Pbx2 does not alter the expression of scapular blade-patterning genes (e.g. Gli3, Alx4). Of note, although we previously reported rostral shifts in Hox gene expression in Pbx1/2Mut paraxial mesoderm (Capellini et al., 2006), no known single or compound Hox mutant displays blade defects. Only Hox5 mutants exhibit rostral blade shifts, with normal blade morphology (Rancourt et al., 1995). Thus, we cannot yet invoke a mechanism whereby loss of Pbx proteins, as Hox co-factors, leads to the blade phenotypes of Pbx compound mutants via altered Hox function.

As in dermomyotome, the expression of Pbx1-3 does not colocalize in somatopleure. However, they do show overlapping expression in limb mesoderm (Fig. 1A,C). Indeed, Pbx1 and Pbx2 show the most extensive overlapping expression in E8.0-10.5 LPM and proximal limbs (Fig. 1C) (Capellini et al., 2006), whereas Pbx1 and Pbx3 colocalize only in proximal-anterior limbs after E9.5 and within somatopleure after E11.0 (Fig. 1) (Di Giacomo et al., 2006). Accordingly, blade defect severity in Pbx1/2Mut significantly exceeds that of Pbx1/3Mut embryos, and is supported by the accentuated reduction in blade-patterning gene expression in Pbx1/2Mut embryos (see below). These findings indicate that Pbx2 is more crucial than Pbx3 in scapular development, probably because it shares greater spatiotemporal domains with Pbx1, or is expressed at higher levels.

Importantly, most girdle and LPM limb derivatives are hypoplastic and malformed in compound Pbx mutants. Pbx1/2Mut and Pbx1/3Mut scapulae also exhibit markedly abnormal head morphologies. Analyses of known molecular markers of the head (Hoxc6) and acromion (Pax1) revealed that they are spatially diffused and reduced in these mutants, although they are still present in domains that lack detectable changes in cell proliferation and apoptosis. Furthermore, as Pax1 and Hoxc6 are reduced in their domains, other mesodermal genes such as Tbx15 are upregulated. These findings corroborate our previous reports on the disruption of Hox expression in Pbx1/2Mut forelimbs (Capellini et al., 2006). For example, we reported the upregulation of Hoxa9 and Hoxd9, genes that are involved in the growth and patterning of the stylopod in Pbx1/2Mut forelimbs (Capellini et al., 2006). Along with spatial expansions of other 3′ Hox genes (i.e. Hoxc6), such data support our finding of potential scapular head/humeral head duplications in these mutants (Fig. 8B). Thus, the hierarchical role of Pbx factors in Hox gene expression and/or their role as Hox co-factors in the LPM are likely mechanisms for assigning positional information to head progenitors.

Lastly, our results show that whereas Pbx1;Pbx2 and Pbx1;Pbx3 mutants exhibit severe blade and head defects, Pbx2;Pbx3 mutants are normal. Therefore, in light of the overlapping functions of Pbx genes in development (Capellini et al., 2006; Capellini et al., 2008; Stankunas et al., 2008), we establish that the dosage of Pbx1 is paramount also in scapula morphogenesis.

In mice, Emx2^–/– embryos do not initiate mesenchymal condensation in pre-blade domains (Pellegrini et al., 2001), and in chick Emx2 functions upstream of Sox9 (Huang et al., 2006). Emx2 has thus been considered to maintain the positional identity of cells fated to form the blade (Bi et al., 2001; Prols et al., 2004). Here, we reveal that Emx2 controls cell proliferation in the pre-blade mesenchyme. Thus, the reduction of Sox9 and absence of the blade observed in Emx2^–/– mutants (Pellegrini et al., 2001) might be partially due to this perturbed cellular behavior.

In Pbx1/2Mut forelimbs, Emx2 expression was severely reduced (Fig. 1B) and Sox9 expression was diminished across the pre-blade region, as shown by OPT (Fig. 3). Similar, although not as extreme, reductions were observed in Pbx1/3Mut domains (Fig. 3). These data suggest that Pbx genes cooperate and genetically regulate Emx2 (and Sox9). Interestingly, unlike Emx2^–/– embryos, Pbx1/2Mut embryos did not exhibit proliferation defects. It remains unclear how Emx2 dosage influences cell proliferation, as Emx2^+/– embryos retain normal cell proliferation (data not shown) and intact blades. Similarly, Emx2, although reduced in Pbx1/2Mut embryos, might still be present at sufficient levels to sustain cell division. Alternatively, the compound absence of all three Pbx genes might be required for the complete loss of Emx2 expression (Fig. 8B) and reduction of cell proliferation.

Pbx genes also control a second pathway involved in blade patterning, which includes Alx1/Alx4, Gli3, and Tbx15, genes that have specific roles in patterning the superior, central, and inferior blade domains, respectively (Fig. 8B) (Kuijper et al., 2005). From E10-11, they become co-expressed with Emx2 in the proximal limb and somatopleure, within a region fated to form the scapula (Huang et al., 2006). A seminal study (Kuijper et al., 2005) demonstrated that blade patterning is sensitive to gene dosage, with the most extreme reductions occurring when Tbx15 loss is coupled with reduced dosage of either at least one gene crucial for inferior blade patterning (i.e. Gli3), or of at least two genes essential for superior blade patterning (e.g. Alx1 and Alx4). We observed that the expression of these genes in Pbx1^–/–, Pbx1/2Mut and Pbx1/3Mut embryos is disrupted proportionally to the severity of blade defects in each respective mutant and that such abnormalities occur without detectable perturbations in cell proliferation or apoptosis (Fig. 8A). Accordingly, at E10.5, the expression of all known blade-patterning genes was reduced to nearly undetectable levels in Pbx1/2Mut, whereas only a subset of these genes (Alx1, Alx4 and Gli3) was altered in Pbx1/3Mut, and, lastly, only Alx1 was modestly reduced in Pbx1^–/– mutants (Fig. 5). Of interest, Pbx1/3Mut embryos additionally exhibit foramina, a phenotype resembling that reported in extra toes (Gli3) (Hui and Joyner, 1993; Johnson, 1967; Schimmang et al., 1992) and polycomb gene M33^–/– (Cbx2 – Mouse Genome Informatics) (Core et al., 1997) mutant mice. Accordingly, Pbx genes have been shown to mediate polycomb expression in axial mesodermal tissues (Capellini et al., 2008).

Our studies reveal that, unlike Pbx1-3, whose domains of genetic interaction include blade and head regions, Pbx1 and Emx2 interaction is restricted to the blade. Compound Pbx1^+/–;Emx2^+/– and Pbx1^–/–;Emx2^+/– (Pbx1/Emx2Mut) mutants exhibit novel blade defects (i.e. proximal indentations and blade bifurcations) that are absent from single Pbx1^+/–, Pbx1^–/–, Emx2^+/– or Emx2^–/– embryos (Fig. 2) and are likely to arise from reductions in Sox9-positive cells along the vertebral border of the blade. These defects are partially mediated by Emx2 control of cell proliferation, as documented in blade-forming domains of Pbx1/Emx2Mut embryos. Although one copy of Emx2 is present in the latter genotype and Pbx1^–/– embryos lack changes in Emx2 expression, decreased levels of Emx2 might be sufficient to cause proliferation defects. Thus, the bifurcation and substantial blade loss in Pbx1/Emx2Mut embryos might be a partial consequence of reduced mesenchymal cell number and condensations.

Importantly, analyses of Pbx1/Emx2Mut embryos also showed that Pbx1 and Emx2 cooperatively control genes upstream of superior blade patterning (Alx1 and Alx4) (Fig. 8B). Concomitantly, they might govern genes patterning other blade domains, as Emx2^–/– and compound Pbx1/2Mut embryos lack Tbx15 expression (Fig. 5; see Fig. S3 in the supplementary material). However, owing to in utero lethality of Pbx1^–/–;Pbx2^+/– mutants, the generation of more complex Pbx1-3;Emx2 genotypes was impossible.

We also uncovered that Pbx1 and Emx2 heterodimerize at specific DNA sequences. Indeed, the Emx2 homeodomain YPWL motif (Bürglin, 1994), which is identical to part of the `hexapeptide' required for Hox-Pbx interaction (reviewed by Moens and Selleri, 2006), is likely to mediate Emx2-Pbx1 heterodimerization. We found that the Emx2-Pbx1 heterodimer preferentially binds the CTTTAATGAT sequence (EP site), within which TGAT represents the Pbx1 half-site (Lu et al., 1995; Van Dijk et al., 1993). Interestingly, the tandem half-site configuration bound by Emx2-Pbx1 differs from that adopted by Hox-Pbx heterodimers, whereby Hox occupies a half-site 3′ to the TGAT Pbx1 binding sequence (Chang et al., 1996). This difference is not due to a TGAT position bias in the oligonucleotide used in site selection because a similar oligonucleotide, containing random nucleotides at five positions 3′ to the TGAT, did not promote Emx2-Pbx1 heterodimer binding (not shown).

Among the aristaless-related genes involved in scapular development, Alx1 has a pre-eminent function (Kuijper et al., 2005). Expression of Alx1, loss-of-function of which results in superior blade defects (Kuijper et al., 2005), was undetectable in Pbx1/2Mut, Pbx1/3Mut and Pbx1/Emx2Mut embryos (Fig. 5A, left). Alx1 loss occurred in the absence of cell proliferation and apoptosis defects in Pbx1/2Mut embryos, whereas it was accompanied by a decrease in cell proliferation in Pbx1/Emx2Mut embryos. Interestingly, in Emx2^–/– embryos, which exhibit blade loss and a marked decrease in cell proliferation, part of the Alx1 expression territory was maintained. Overall, these data strongly suggest that Pbx and Emx2 factors are required for Alx1 transcription. Indeed, we identified a conserved element (Alx1 5′-7) upstream of the putative Alx1 TSS that is bound in vivo by Emx2-Pbx1, suggesting that both proteins directly control Alx1 expression in pre-scapular domains. Although the possibility cannot be excluded that other sites more distant from Alx1 might recruit Emx2 and/or Pbx1, we did not detect additional binding by Emx2-Pbx1 within the 16 evolutionarily conserved regions that were analyzed, covering more than 25 kb around Alx1. Interestingly, the Alx1 5′-7 element lacks a DNA sequence identical to the EP site, but displays two nearby conserved motifs that are both required for cooperative Emx2-Pbx1 binding. Of these, one contains the TGAT Pbx half-site (Site A), and the other a TCAT motif (Site B), which is also present within the selected EP site (complementary strand). Additionally, other predicted Pbx1 and Emx2 sites within the conserved element are not bound by their respective proteins as assayed by EMSA. The Alx1-EPBS element furthermore mediated transactivation in transient transfections of cultured cells only when Emx2 and Pbx1 were co-expressed. Interestingly, whereas mutation of the TGAT (Site A) or the TCAT (Site B) motif alone within the Alx1-EPBS element caused a discrete decrease in Emx2-Pbx1-mediated transactivation, only the compound mutation of both motifs significantly reduced reporter activity, underscoring that both Sites A and B are necessary for full transcriptional activation by Emx2-Pbx1.

In conclusion, here we have addressed the hierarchical control of scapula development by Pbx genes and their interaction with Emx2. We have uncovered that Pbx genes act upstream of head development and control genetic pathways underlying blade formation (Fig. 8B) (Kuijper et al., 2005). Importantly, we have established that Pbx genes hierarchically control Emx2 and blade mesenchymal condensation, and that Pbx1 and Emx2 genetically and physically interact in blade patterning. Lastly, by highlighting a direct synergistic regulation of Alx1 by Pbx1-Emx2, we have delineated a novel genetic network underlying shoulder girdle development.

We are grateful to Matt Koss for apoptosis assays; Michael Cleary for anti-Pbx antibodies; Mouse Genetics Community researchers for in situ probes; and Liz Lacy and Ann Foley for discussions. T.D.C. was the recipient of the CUNY Carole and Morton Olshan Fellowship. Work was supported by grants from the NIH (2R01HD043997-06 [ARRA Suppl], 3R21DE018031-02S1 and 1R01HD061403-01 to L.S.); March of Dimes and Birth Defects Foundation (#6-FY03-071 to L.S.); and the Italian Telethon, ASM Foundation, and the Italian Association for Cancer Research to V.Z. L.S. is an Irma T. Hirschl Scholar, a recipient of grants from The Alice Bohmfalk Trust and The Frueauff Foundation, and a recipient of a research award for `Cleft Lip/Palate and Craniofacial Anomalies' from the Cleft Palate Foundation. Deposited in PMC for release after 12 months.