Decoupling the function of Hox and Shh in developing limb reveals multiple inputs of Hox genes on limb growth

One of the most intriguing issues in developmental biology is how organ growth and patterning are coordinated during embryogenesis. For example, in vertebrates, the formation of the main body plan is achieved through posterior elongation, and cells ingressing through the primitive streak progressively express more posterior identity genes. Similarly, limb development relies on an exquisite coordination between growth and patterning, but the underlying mechanisms remain elusive. Limb outgrowth requires FGF signaling from the AER (Fallon et al., 1994; Niswander and Martin, 1993), provided by Fgf4, Fgf8, Fgf9 and Fgf17 (collectively called AER-FGFs). Among them, Fgf8 is the only FGF expressed from nascent AER until AER regression and is sufficient to ensure virtually normal limb development in absence of the other AER-FGFs (Boulet et al., 2004; Moon et al., 2000; Sun et al., 2000; Sun et al., 2002). Three mesenchymal factors, Fgf10, Shh and the BMP antagonist Grem1 are essential for correct expression of AER-FGFs (Laufer et al., 1994; Ohuchi et al., 1997; Zúñiga et al., 1999). Initially, in the presumptive limb field of the lateral plate mesoderm (LPM), Fgf10 triggers the expression of Fgf8 in the overlaying ectoderm, which in turn maintains Fgf10 expression in the early limb bud (Ohuchi et al., 1997). Subsequently, further growth along the proximodistal (PD) axis and the anteroposterior (AP) expansion of the distal bud (presumptive hand/foot) is driven by a positive regulatory feedback loop between Shh, Grem1 and AER-FGFs, with Shh signaling being required to maintain Grem1 transcription while the latter preserves expression of AER-FGFs by antagonizing BMPs (e.g. Zeller et al., 2009). In mice, expression of Shh and AER-FGFs regresses around E13 and subsequent growth of the limb relies on the elongation of skeletal elements.

The combined inactivation of the HoxA and HoxD gene clusters previously revealed their requirement for the activation and maintenance of Shh expression (Kmita et al., 2005). These genes are members of the Hox family, which encode transcription factors that control patterning events during embryogenesis. In most vertebrates, Hox genes are grouped into four clusters (HoxA to HoxD) in which the relative order of the genes on the chromosome parallels their serial expression in time and space, a phenomenon referred to as temporal and spatial co-linearity (Duboule and Morata, 1994). For example, in nascent limb buds, genes from the HoxA and HoxD clusters are activated sequentially in time (from group 1 to group 13) and their expression patterns switch from uniform for early activated genes to posteriorly restricted for later activated ones (reviewed by Zakany and Duboule, 2007). Interestingly, only those expressed in the posterior limb mesenchyme (group 10 to 13) are able to activate Shh in vivo (Tarchini et al., 2006) via direct activation of the Shh limb enhancer (Capellini et al., 2006), thereby establishing a link between the co-linear Hox activation and Shh-dependent AP polarity (Tarchini et al., 2006). Accordingly, ectopic/precocious expression of Hoxd13 and Hoxd12 anteriorly results in mirror-image expression of Shh and bilaterally symmetrical limbs (Knezevic et al., 1997; Zákány et al., 2004). Recent studies showed that the activation of Shh actually relies on a Hox-Hand2 protein complex (Galli et al., 2010) and Hand2 itself is activated by paralogous group 9 Hox genes (Xu and Wellik, 2011). Together, these results show both direct and indirect control of Shh expression by specific members of the Hox family of transcription factors. In turn, Shh signaling is required for the maintenance of HoxA and HoxD gene expression by inhibiting the proteolytic cleavage of the Gli3 protein into a transcriptional repressor (Litingtung et al., 2002; te Welscher et al., 2002).

Inactivation of the HoxA and HoxD clusters eventually results in severe reduction of the limb size, which was initially associated with Shh downregulation (Kmita et al., 2005). However, the respective role of Shh and HoxA;D genes has remained unclear owing to the Hox-Shh positive cross-regulation. In addition, Shh^-/- limb shortening is significantly less severe compared with the double HoxA;HoxD mutant (Chiang et al., 2001; Kmita et al., 2005; Kraus et al., 2001) suggesting that HoxA and HoxD genes also impact on limb growth independently of their control of Shh expression. Here, we report on the Hox-dependent limb growth in contexts where Hox and Shh functions are uncoupled. First, by investigating HoxA;D function before the maintenance of AER-FGFs becomes associated with Shh signaling. Second, by studying the role of HoxA;D genes in absence of Gli3, which renders Shh functionally irrelevant.

Mutant mouse lines have been described previously: floxed HoxA gene cluster (Kmita et al., 2005), HoxAnull (Scotti and Kmita, 2012), HoxDnull (Spitz et al., 2001) [also referred to as HoxD^- and named TgHd11/lacZDel9 by Spitz et al. (Spitz et al., 2001)], Gli3XtJ (Hui and Joyner, 1993) and Prx1Cre (Logan et al., 2002). The Shhnull line was derived from crossing the conditional Shh mutant (Lewis et al., 2001) with Mox2Cre mice (Tallquist and Soriano, 2000). Mice and embryos were genotyped by PCR or Southern blot analysis, using genomic DNA extracted from tail biopsies and yolk sacs, respectively.

Optical projection tomography (OPT) microscopy (Sharpe et al., 2002) was performed according to manufacturer specifications. Briefly, stained forelimbs skeletons or buds were embedded in 1% low-melt agarose, then dehydrated in 100% methanol and cleared in a mix of Benzyl Alcohol and Benzyl Benzoate (1:2). Scanning was performed using the Bioptonics 3001M OPT Scanner (Bioptonics, UK) with SKYSCAN-3001 (Skyscan, Belgium). Three-dimensional OPT reconstructions were performed with NRecon software (Skyscan) and visualized with Bioptonics viewer (Bioptonics). Measurements were made using Imaris (Bit plane).

Whole-mount in situ hybridizations were performed according to standard procedures using the previously described Shh (Echelard et al., 1993), Fgf8 (Crossley and Martin, 1995), Grem1 (Zúñiga et al., 1999), Bmp4 (Bénazet et al., 2009), Hoxa11 and Hoxa13 (Warot et al., 1997) antisense riboprobes.

P19 cells were transfected with 250 ng of the luciferase reporter under the control of the minimal promoter with or without enhancer sequences together with 100 ng of either Hoxd9- or Hoxa13-coding sequence or control-GFP DNA. Each plasmid combination was set in triplicates. Cell media were changed 24 hours after transfection and cells were kept in incubation for an additional 24 hours. Cell media was discarded and lysis buffer [100 mM Tris (pH 7.8), 0.5% NP-40, 1 mM DTT] was placed onto cells. Cell lysate was harvested and placed in contact with luciferin (Promega) in order to measure the reporter activity using Promega’s GLOMAX 96 microplate luminometer. Fold activation was measured by comparing mean RLU values for each condition with the mean RLU value obtained for the condition determined as basal level, i.e. control luciferase reporter (no enhancer sequence) in presence of control-GFP DNA.

Limb growth initially relies on interactions between the AER and the mesenchyme (referred to as the AER phase hereafter) until the AER regresses. Subsequently, elongation of skeletal elements defines the eventual length of the limb (referred to as the post-AER phase hereafter). We first examined whether the impact of HoxA;D function on the mesenchyme-AER interactions is mediated only by the regulation of Shh expression or whether it includes an independent effect of HoxA;D genes on the other factors implicated in mesenchyme-AER interactions. As HoxA;D function is required for the initial activation of Shh transcription, there is a narrow time window, at early bud stage, when HoxA and HoxD genes are functional, while Shh signaling is not yet operational. We thus took advantage of this Shh-independent stage to assess whether the lack of HoxA;D genes interferes with the expression of factors known to maintain AER-FGFs. Whole-mount in situ hybridization shows readily detectable Grem1 expression in wild-type (Fig. 1A) and Shh^-/- (Fig. 1B) nascent buds. Grem1 expression is actually still detectable in Shh^-/- buds at E10 (Fig. 1C), consistent with previous results showing that Shh is dispensable for the transcriptional activation of Grem1 (Panman et al., 2006; Bénazet et al., 2009). By marked contrast, the ubiquitous deletion of the HoxA and HoxD clusters results in the complete lack of Grem1 activation in early limb buds (Fig. 1D), indicating that the initial activation of Grem1 requires HoxA;D function. Previous work has revealed that bone morphogenetic protein 4 (Bmp4) is required for the transcriptional activation of Grem1 in early limb buds (Bénazet et al., 2009), raising the possibility that the absence of Grem1 expression in Hox(A;D)null mutants is secondary to Bmp4 downregulation. However, Bmp4 remains expressed in Hox(A;D)null limbs (supplementary material Fig. S1), thus excluding its downregulation as a cause for the lack of Grem1 expression. In turn, these results reveal that Bmp4 signaling is unable to trigger Grem1 activation in the absence of HoxA;D genes. To further test the competence of Hox proteins in activating Grem1, we next performed transcription assays in P19 cells, in which the luciferase reporter was under the control of the previously identified Grem1 limb enhancer (Zuniga et al., 2012). In these assays, a plasmid encoding Hoxd9 was used as a representative of Hox proteins present in the nascent limb bud. Luciferase quantification reveals robust increase in the activity of the Grem1 enhancer in presence of the Hoxd9 protein (Fig. 1E), indicating that Hoxd9 (and most likely the other early Hox genes products) is able to activate Grem1 expression. This result, however, does not exclude the requirement of Bmp4 together with HoxA;D genes for Grem1 activation, as Bmp4 is expressed in P19 cells (Chang et al., 2010).

Up to E10, the maintenance of Fgf8 expression in the AER does not require Grem1 function, as revealed by the analysis of Grem1-deficient limb buds (Michos et al., 2004). Consequently, the Fgf8 downregulation observed in Hox(A;D)null nascent buds (Fig. 1D) suggests that another mesenchymal factor maintaining Fgf8 expression is impaired in these mutant buds. We thus analyzed the expression of Fgf10, the function of which triggers Fgf8 activation and contributes to Fgf8 maintenance (Ohuchi et al., 1997). Our results show that Fgf10 is downregulated anteriorly in Hox(A;D)null buds (Fig. 2C) compared with both wild-type (Fig. 2A) and Shh^-/- buds (Fig. 2B), suggesting that HoxA;D genes contribute to Fgf10 regulation in early buds. As Fgf10 is initially activated in the lateral plate mesoderm, we also checked Fgf10 expression prior to limb budding but failed to assess unambiguously whether its expression is affected or not in the presumptive limb field (not shown). In order to have an independent assessment of the capacity of Hox proteins to positively regulate Fgf10 expression, we performed transcription assays in P19 cells. To drive expression of the luciferase reporter, we used a DNA fragment previously identified as containing the Fgf10 limb enhancer (R2 in Fig. 2D,E) (Ohuchi et al., 2005; Sasak et al., 2002). As negative controls, we linked the reporter to the other conserved DNA fragments located upstream Fgf10 but with no enhancer function in limbs (R1 and R3 in Fig. 2D,E), one of them acting as Fgf10 enhancer in the inner ear [R1; see also Ohuchi et al. (Ohuchi et al., 2005)]. Upon co-transfection with the Hoxd9-encoding plasmid, there is a significant augmentation of the R2 enhancer activity, as revealed by the reporter expression, whereas the activity of R1 and R3 remained at basal levels (Fig. 2E). These results point to a specific effect of Hoxd9 protein on the activity of the enhancer controlling Fgf10 expression in nascent buds and support the contribution of HoxA;D genes for the proper expression of Fgf10 in the limb mesenchyme.

The requirement of Hox function for Grem1 transcriptional activation raises the possibility that HoxA;D genes are also involved in Grem1 regulation at later stages. Investigating this hypothesis necessitates the inactivation of HoxA;D genes after Grem1 initial activation. This can be achieved using the HoxAc/-;D^-/- conditional mutant, in which transcripts from the conditional HoxA allele are transiently generated (up to E10.5) owing to the kinetics of the Prx1-Cre activity (Kmita et al., 2005). Consistent with the residual HoxA transcription, Grem1 is activated in HoxAc/-;D^-/- buds (supplementary material Fig. S2). At E11.5, Grem1 expression remains detectable in HoxAc/-;D^-/- mutants, although it fails to expand anteriorly as it does in wild-type buds (Fig. 3A,B). Interestingly, Fgf8 expression, which is normal at E10.5 (supplementary material Fig. S2C), is subsequently restricted to the part of the AER that parallels the Grem1 domain along the AP axis (Fig. 3E, arrows), suggesting that, at this stage, Grem1 defines the AP domain of Fgf8 expression in the AER.

Although Grem1 expression is posteriorly restricted in the HoxAc/-;D^-/- mutant, the respective role of Hox genes and Shh in Grem1 anterior expansion remains unclear because Shh is downregulated in HoxAc/-;D^-/- limb buds (Kmita et al., 2005). To investigate the Shh-independent function of HoxA;D genes, we needed to remove Gli3 function to render Shh functionally irrelevant (Litingtung et al., 2002; te Welscher et al., 2002). Owing to the lethality associated with Gli3 inactivation and HoxAc/-mutation as well as the infertility of HoxD^-/- mice, triple mutant embryos can only be generated from crosses between triple heterozygous mice. However, HoxA^+/-;HoxD^+/-;Gli3^+/- turns out to be poorly fertile, which drastically reduces the feasibility of studying HoxAc/-;HoxD^-/-;Gli3^-/- embryos. We could circumvent this limitation using the conditional HoxA allele instead of the null allele and we therefore performed our analysis with triple mutants in which both HoxA alleles are conditional (HoxAc/c;HoxD^-/-;Gli3^-/-), which were obtained at the expected Mendelian ratio (1/128). The difference between HoxAc/c and HoxAc/-is that the completion of Cre-mediated deletion is delayed with two floxed alleles and some HoxA expression remains at least up to E11.5 (supplementary material Fig. S3). Comparison of HoxAc/c;HoxD^-/-;Gli3^-/- with HoxAc/c;HoxD^-/- buds, at E11.5, reveals an anteriorization of both Grem1 and Fgf8 in the triple mutant (Fig. 3C-D). However, expression of both genes remains clearly impaired compared with Gli3^-/- and Shh^-/-; Gli3^-/- mutant (Fig. 3E) (see also Litingtung et al., 2002; te Welscher et al., 2002). Notably, in HoxAc/c;HoxD^-/-;Gli3^-/- buds, Grem1 expression is mosaic and significantly reduced in the anterior mesenchyme (Fig. 3D compared with 3E). This indicates that Gli3 inactivation is not sufficient to restore the anterior expansion of Grem1 in HoxAc/c;HoxD^-/-;Gli3^-/- as it does in Shh^-/-;Gli3^-/- buds and raises the possibility that HoxA;D genes contribute to this process.

We reasoned that if Hox proteins contribute to Grem1 anterior expansion, Grem1 expression in HoxAc/c;HoxD^-/-;Gli3^-/- should be, at least in part, due to the residual transcription of the conditional HoxA allele, notably that of Hoxa13, which is the most distally expressed HoxA gene. Accordingly, the gain of Grem1 upon Gli3 inactivation (Fig. 3D) coincides with the anterior expansion of the residual Hoxa13 expression (supplementary material Fig. S4B), which itself is due to the lack of the Gli3R repressor. Grem1 expansion could also be secondary to the lack of Gli3R. However, the correlation between the mosaicism of Grem1 expression and that of the residual Hoxa13 expression observed in HoxAc/c;HoxD^-/- and HoxAc/c;HoxD^-/-;Gli3^-/- (Fig. 3C,D; supplementary material Fig. S5) strongly favors the interpretation that HoxA;D genes are required for the proper anterior expansion of Grem1. Accordingly, a single wild-type allele of the HoxA or HoxD cluster is sufficient to recover the anterior Grem1 expression as in Gli3^-/- and Shh^-/-;Gli3^-/- (shown for the HoxD wild-type allele in Fig. 3F, compare with 3D and 3E).

To test independently the effect of the Hoxa13 protein on Grem1 expression, we performed a transcription assay in P19 cells. Co-transfection of the expression vector encoding for Hoxa13 together with the Grem1 limb enhancer linked to the luciferase reporter resulted in a significantly higher reporter expression, as revealed by luciferase quantification (supplementary material Fig. S6), confirming that Hoxa13 has a positive impact on the activity of the Grem1 limb enhancer. Together, these results provide evidence that HoxA;D genes contribute to the anterior expansion of Grem1 expression independently of their function in controlling Shh.

Our data on Grem1 expression support the notion that HoxA;D genes promote limb bud growth in parallel to their function in regulating Shh. In turn, it suggests that limb bud growth should vary according to the HoxA;D dose, independently of Shh expression. We thus performed measurements of limb buds using optical projection tomography (OPT) imaging to characterize how reducing the HoxA;D gene dose in the Gli3^-/- background affects limb bud growth. At E11.5, reduction of the Hox dose to a single HoxA or HoxD wild-type allele results in the shortening of the limb bud along the PD axis (Fig. 4A,B). In addition, this shortening is more pronounced upon further reduction of the HoxA;D dose (Fig. 4C,D). These data reveal a tight link between the HoxA;D dose and the extent of PD bud growth in the Gli3 mutant background. Our series of HoxA;D;Gli3 mutants also show that one functional allele of either HoxA or HoxD is sufficient to generate the distal ‘fan-shaped’ characteristic of Gli3^-/- limb bud (Fig. 4A,B) but this shape is lost with lower Hox doses (Fig. 4C,D). This suggests that the AP expansion of the distal bud observed in Gli3^-/- and Shh^-/-;Gli3^-/- buds is associated with the function of HoxA;D genes.

Together, our series of HoxA;D;Gli3 mutants reveals a link between the HoxA;D dose and the extent of PD and AP bud growth in the Gli3^-/- background.

In absence of Shh, the domain of Gli3R, which is normally restricted to the most anterior part of the limb bud, expands towards the posterior part (Wang et al., 2000). This results in the downregulation of several genes in most of the Shh^-/- limb bud, including HoxA;D genes (Chiang et al., 2001; Kraus et al., 2001). The rescue of the Shh^-/- limb phenotype upon removal of Gli3 indicates that the downregulation of the direct and/or indirect targets of Gli3R is responsible for the Shh^-/- phenotype. Our finding that the HoxA;D gene dose in the Gli3^-/- background correlates with limb bud growth thus suggests that Shh^-/- limb truncation is, at least in part, the consequence of Gli3R-mediated HoxA;D downregulation. Nonetheless, as Shh remains transiently expressed in the conditional HoxA;D mutant (Kmita et al., 2005), there is less Gli3R-mediated gene repression, including less repression of the residual HoxA expression, than in the Shh mutant and thus it is possible that HoxAc/c;HoxD^-/- buds could develop better than Shh^-/- buds until complete HoxA inactivation. In agreement with this view, at E11.5, Shh^-/- buds are smaller than HoxAc/c;HoxD^-/- buds (Fig. 4E,F).

The size difference between Shh^-/- and the conditional HoxA;D buds is opposite when skeletons are compared (Fig. 4G,H), indicating that the elongation of the skeletal elements is favored in Shh mutant. Interestingly, around E12, there is a complete deficiency of Hox expression in HoxAc/c;HoxD-/-buds, while Hox expression persists in Shh mutant (Fig. 4I-J). Actually HoxA11 expression in Shh mutant appears even stronger at E12.5 than at E11.5 (Fig. 4I; supplementary material Fig. S7). This sustained Hox expression in Shh^-/- (Fig. 4I) could explain the favored elongation of skeletal elements, consistent with previous reports proposing that the elongation of skeletal elements requires the function of HoxA;D genes (Boulet and Capecchi, 2004; Gross et al., 2012; Zákány et al., 1997). In Shh^-/-;Gli3^-/- limb, proper elongation of skeletal elements is thus expected to be associated with HoxA;D function. Yet it remains unclear whether other Gli3R targets are implicated. To clarify this issue, we compared the skeleton of the conditional HoxA;D and HoxA;D;Gli3 limbs and found that they are similarly shorten along the PD axis (Fig. 4K-L). This result indicates that among Gli3R targets, HoxA;D genes have the most relevant role for the elongation of skeletal elements.

The HoxAc/c;HoxD^-/-;Gli3^-/- skeleton at E16.5 is characterized by the increased number of distal rays when compared with HoxAc/c;HoxD^-/- skeletons (Fig. 4K,L). This increased number of distal rays is reminiscent of the increased number of digits in Gli3^-/- limbs, thus raising the possibility that these distal rays correspond to digit remnants. However, the lack of digit-specific features, such as segmentation into distinct phalanges, makes it difficult to assign an identity unambiguously to these rays. Interestingly, HoxAc/+;HoxD^-/- and HoxAc/c;HoxD^+/- limbs, although being significantly smaller than wild-type ones, have the three domains (stylopod, zeugopod and autopod), allowing for unambiguous identification of digits despite their abnormal shape and absence of segmentation into phalanges (Fig. 5A,C). Moreover, the comparison with HoxAc/+;HoxD^-/-;Gli3^-/- and with HoxAc/c;HoxD^+/-;Gli3^-/- limbs (Fig. 5B,D) shows that Gli3 inactivation in these Hox mutant backgrounds primarily affects the number of digital condensations. Together, these data support our interpretation that the distal rays in HoxAc/c;HoxD^-/-;Gli3^-/- limbs are digit remnants.

Interestingly, the bending systematically found in the middle of HoxAc/c;HoxD^-/- skeletons (arrow in Fig. 4H) evokes the bending of wild-type limbs at the transition between stylopod and zeugopod (Fig. 5E). Remarkably, Shh mutant embryos form either a clearly distinct stylopod and zeugopod separated by a rudimentary elbow joint (e.g. Litingtung et al., 2002) or limbs with only the bend typically associated with the elbow joint (arrow in Fig. 4G) (see also Zhu et al., 2008), supporting that the bending observed in HoxAc/c;HoxD^-/- mutant limbs corresponds to a vestige of the stylopod-zeugopod boundary. Thus, the phenotype of HoxAc/c;HoxD^-/- and HoxAc/c;HoxD^-/-;Gli3^-/- skeletons suggests that the transient expression of HoxA genes is sufficient to provide cells with a stylopod, zeugopod or digit identity depending of which HoxA genes they express.

During limb development, genes of the HoxA and HoxD clusters are expressed in a spatial and temporal co-linear manner. Genetics studies previously established the role of Hox genes in limb skeletal patterning but their role in limb bud growth remains unclear, mainly owing to the crossregulation existing between HoxA;D genes and Shh. The work reported here provides a novel perspective on the role of Hox genes in limb bud growth and the mechanisms involved. First, we show that in nascent bud, prior to the normal activation of Shh, the function of HoxA;D genes is required for correct Fgf8 expression. Moreover, we establish that HoxA;D genes achieve this previously unrecognized role through the transcriptional activation of Grem1 and proper expression of Fgf10 in the mesenchyme. Second, taking advantage of Gli3 inactivation that renders Shh expression functionally irrelevant, we provide evidence that the anterior expansion of Grem1 expression, and thus the propagation of Grem1-mediated maintenance of AER-FGFs, requires the function of HoxA;D genes independently of their activity in regulating Shh expression. Accordingly, our data show that during the Shh functional phase, the proper AP and PD growth of the limb bud requires the Shh-independent functions of HoxA;D genes. Finally, we also show that HoxA;D genes are subsequently required for the elongation of skeletal elements.

Limb bud growth first relies on epithelial-mesenchymal (EM) interactions that trigger and then maintain expression of AER-FGFs, up to AER regression around E13. The initial EM interaction, which induces limb budding, relies on Fgf10 signaling from the lateral plate mesoderm (LPM) that triggers expression of Fgf8 in the overlaying ectoderm (Ohuchi et al., 1997). Activation of Fgf10 expression in the presumptive forelimb field of the LPM relies on Tbx5 (Agarwal et al., 2003; Rallis et al., 2003). Recently, analysis of Tbx5 regulation identified Hox proteins from paralogous group 4 and 5 as regulators of Tbx5 expression in the LPM (Minguillon et al., 2012). Despite the absence of Hoxd4, Hoxa4 and Hoxa5 in the LPM of Hox(A;D)null embryos, we did not detect clear modification of Fgf10 expression in the LPM (not shown), suggesting that the absence of HoxA;D genes in the LPM has minor, if any, impact on the Tbx5-mediated regulation of Fgf10. Therefore, the effect of the Hox(A;D)null mutation on Fgf10 expression is primarily associated with the function of HoxA;D genes in the developing limb bud, consistent with the evidence that Hoxd9 protein positively influence the activity of the Fgf10 early limb enhancer in cell culture assays. The evidence that Fgf10 expression is not abrogated in Hox(A;D)null limb bud but only reduced suggests that Fgf10 regulation involves HoxA;D proteins, as well as other transcription factors. Interestingly, switching expression of Hoxd12 and Hoxd13 from late to early bud stages in absence of Gli3 suppresses Fgf10 expression in the limb mesenchyme (Zakany et al., 2007). It was thus proposed that Fgf10 downregulation is the consequence of the functional suppression of early expressed Hox by ‘late’-activated Hox genes (Zakany et al., 2007), which supports the model that Hox genes expressed in early limb buds act as positive regulators of Fgf10. Together, these results point to the implication of Hox genes in the early EM interactions, first through the impact of both group 4 and 5 Hox genes on Fgf10 expression in the LPM and second through the function of HoxA;D genes in the nascent limb bud.

Although Grem1 is activated in nascent bud, its function in maintaining Fgf8 in the AER becomes indispensable around E10, as revealed by the analysis of the Grem1 mutant (Michos et al., 2004). Grem1 downregulation in Shh mutant and the ability of Shh-soaked beads implanted in the anterior mesenchyme to trigger ectopic Grem1 expression provided evidence that Shh signaling acts positively on Grem1 expression (Zúñiga et al., 1999). However, Shh signaling is not needed for the initial activation of Grem1 in nascent bud (Panman et al., 2006; Bénazet et al., 2009). Moreover, inactivating Shh together with its direct intracellular mediator Gli3 prevents Grem1 downregulation (Litingtung et al., 2002; te Welscher et al., 2002), indicating that Grem1 downregulation in Shh mutant is due to Gli3R activity, although Gli3 repression of Grem1 has been proposed to be indirect (Vokes et al., 2008). As Gli3R also leads to HoxA;D downregulation, it is possible that Gli3R repression of Grem1 is secondary to Hox repression. Our results strongly support this interpretation: first, in contrast to Shh, HoxA;D function is required for the initial activation of Grem1 in nascent bud. Second, Gli3 inactivation in the conditional HoxA;D mutant background is not sufficient to restore Grem1 expression, as reported for the Shh^-/-;Gli3^-/- mutant. Finally, the mosaicism of Grem1 in the conditional HoxA;D;Gli3 mutant strongly correlates with the mosaicism of the residual HoxA gene expression in this mutant.

Based on this previously unappreciated function of HoxA;D genes in eliciting Grem1 expression, which occurs independently of their role in regulating Shh expression, we propose a novel model for the control of Grem1 expression (Fig. 6). In this model, HoxA;D genes trigger the initial activation of Grem1 (either in cooperation or in parallel with Bmp4 signaling) in nascent limb bud. Concomitantly, they activate Shh expression in the ZPA. In turn, Shh signaling restricts Gli3 proteolytic cleavage into a transcriptional repressor to the most anterior mesenchymal cells, thereby defining a domain permissive for the expression of HoxA;D genes and subsequent propagation of Grem1 expression.

By uncoupling Hox and Shh functions using Gli3 inactivation, which renders Shh functionally irrelevant, we now provide evidence that Hox genes control bud growth in parallel to their activity in controlling Shh expression. Notably, we found that during the AER phase, reducing the HoxA;D gene dose in absence of Gli3 interferes with limb bud growth in a dose-dependent manner, both along the AP and PD axes. Previous studies suggesting that the elongation of skeletal elements requires HoxA;D function were based on ubiquitous gene inactivation (Boulet and Capecchi, 2004; Gross et al., 2012; Zákány et al., 1997). Our data suggest that at least part of the reported phenotype could have been due to defective growth during the AER phase. Nonetheless, the comparison between the size of Shh^-/- and the conditional HoxA;D mutant at E11.5 and E16.5 further supports the requirement of Hox products for the elongation of the skeletal elements at the post-AER stages. Moreover, the difference in Fgf8 expression in the conditional HoxA;D and HoxA;D;Gli3 mutants (and thus the difference in the size of the pool of skeletal precursors) has no detectable effect on the eventual size of the skeleton along the PD axis, further highlighting the importance of HoxA;D function in the elongation of skeletal elements.

Together, these results provide compelling evidence for a Hox-dependent Shh-independent mechanism that promotes limb bud growth during the AER phase, as well as the requirement of HoxA;D genes for the subsequent elongation of skeletal elements.

Exhaustive analysis of the function of AER-FGFs suggested that AER-FGF signaling contributes to PD patterning in addition to limb bud growth (Mariani et al., 2008). Interestingly, recent work provided evidence that AER-FGF signaling has a key role for the expression of ‘late/distal’ Hox genes (i.e. group 11 to 13), as revealed by the Fgf8 requirement for the sequential activation of Hoxa11 and Hoxa13 in limb mesenchymal cells in culture (Cooper et al., 2011; Roselló-Díez et al., 2011) and Hoxd13 downregulation associated with a reduced dose of AER-FGFs (Sun et al., 2002). These genes actually instruct cells of their PD identity (zeugopod identity for Hox11 and autopod identity for Hox13) and thereby likely represent the effectors through which AER-FGFs influence PD patterning. Our finding that early activated HoxA;D genes promote the maintenance of AER-FGFs suggests that, in addition to providing proximal identity to cells where they are expressed, early activated HoxA;D genes ensure proper AER-FGF-dependent expression of late/distal HoxA;D genes. Accordingly, the Hox-FGF-positive cross-regulation would control limb bud growth and ensure some steadiness between the formation of proximal and distal identity.

In summary, our data uncover a previously unappreciated role of HoxA;D genes in the control of AER-FGFs expression, which provides a novel perspective on the role of HoxA;D genes in limb growth (Fig. 6). First, in nascent limb buds, early-activated HoxA and HoxD genes contribute to the proper expression of Fgf10 and as a consequence proper Fgf8 expression in the AER. Concomitantly, Hox proteins are required to activate Grem1 and Shh and set up the cross-regulation between Shh, Gli3 and HoxA;D genes as well as the Grem1-dependent maintenance of AER-FGFs. In turn, AER-FGF signaling subsequently triggers the sequential activation of ‘late’ HoxA and HoxD genes, which specify intermediate (zeugopod) and distal (autopod) identity and are essential to sustain growth along both P-D and A-P axis. In parallel, Shh signaling, by restricting the processing of the Gli3 protein into a transcriptional repressor to anterior limb mesenchymal cells, maintains the A-P biased expression of HoxD genes that subsequently translates into A-P patterning. The dual role of HoxA;D genes in controlling growth pathways and patterning, from early limb bud stages onwards, establishes a molecular link between these two processes, which need to be precisely coordinated to ensure robustness of the limb architecture.

We are thankful to Aurélie de Ronne, Jessica Barthe and Marisa Junco for technical help, and to Pierre Chambon, Denis Duboule and Rolf Zeller for providing mutant lines. We are particularly grateful to Rolf Zeller and Cliff Tabin for insightful comments on our data.