Regulatory integration of Hox factor activity with T-box factors in limb development

Tetrapod limbs develop under the control of a common genetic program that is conserved across species (Schneider and Shubin, 2013; Sears et al., 2015). Limb type-specific transcription factors (TFs) modulate this common program throughout limb development, playing crucial roles in each limb as early as limb bud initiation. In mice, the forelimb-restricted T-box TF Tbx5 is required for initiation of forelimb (FL) bud development through control of Fgf10 expression, while Tbx4, a paralog of Tbx5 with hindlimb-enriched expression, plays a role in early hindlimb (HL) bud formation (Agarwal et al., 2003; Naiche and Papaioannou, 2003; Rallis et al., 2003). These two TFs also play a role in conferring limb type-specific features (Hasson et al., 2010; Ouimette et al., 2010).

Recent work showed a high degree of similarity between the developmental programs that drive FL and HL development at both the epigenetic and transcriptomic levels (Cotney et al., 2012; Nemec et al., 2017). Pitx1, a TF with HL-restricted expression, is crucial for HL development and is an upstream regulator of Tbx4 (Lanctôt et al., 1999; Logan and Tabin, 1999). Global analysis of the Pitx1 gene regulatory network showed that Pitx1 implements limb type-specific features by its action on diverse genomic targets common to both FL and HL (Nemec et al., 2017). This mode of action is likely to apply to other limb-restricted TFs in light of the extensive similarities between the FL and HL programs.

Hox genes play a crucial role in establishment of the anterior-posterior (AP) body axis (Wellik, 2007). Two waves of Hox TF expression first control limb proximodistal (PD) development and, second, AP patterning in both FL and HL (Zakany and Duboule, 2007; Zakany et al., 2004). Limb expression of Hox and of Tbx genes are thus necessarily overlapping (Fig. 1A), with complex patterns that vary both in space and time. Loss-of-function mouse mutants for Tbx4 and Hox genes support the idea that their actions on the limb program may intersect. For example, loss of Tbx4 early in HL development leads to severe HL bud malformations, whereas conditional loss of Tbx4 at later stages causes deformities, such as bone growth defects in the proximal limb (femur and fibula), anterior digit fusions in the autopod, and muscle connective tissue patterning defects (Fig. 1B) (Hasson et al., 2010; Naiche and Papaioannou, 2007). Loss of Hox genes affects the developing limb in a region-specific manner: loss of Hox10 paralogs leads to hypoplasia in the stylopod, while loss of Hox11 paralogs results in malformed zeugopod structures, and Hox13 paralogs are necessary for autopod development (Fig. 1B) (Fromental-Ramain et al., 1996; Raines et al., 2015; Wellik and Capecchi, 2003). In this context, it is noteworthy that inactivation of the Hoxc10 gene, which is only expressed in HL but not FL, does not produce a phenotype on its own, suggesting redundancy between subsets of Hox genes. Further, cooperation between Hox genes and limb-restricted TFs such as Tbx4 and Tbx5 may provide a mechanism to increase specificity and diversity of action.

In the present work, we first addressed this putative cooperation by defining the putative targets of action of the HL-restricted Tbx4 and Hoxc10 using ChIPseq, and then by investigating mechanisms of their joint action. The data reveal a large proportion of common genomic targets for these two factors, many of which harbor a composite binding site for Tbx4 and Hoxc10. The joint action of these factors leads to synergistic activation of transcription. Both factors are necessary for action on the composite target, and Hoxc10 behaves as a modulator of Tbx4-dependent activity. A large proportion of common Tbx4/Hoxc10 targets also bind Pitx1 in HL. In contrast to Hoxc10, Hoxd13 also interacts with Tbx4 (and Tbx5) but with the opposite outcome, antagonizing its activity. As Hox13 paralogs are deemed to play similar roles in FL and HL, we compared genomic occupancy of Tbx5 in FL with that of Tbx4 in HL, and again found widespread overlap in their putative enhancer targets that are in an active chromatin state. In addition, Hoxd13 interacts with Tbx5 to antagonize its activity. Collectively, this work reveals a mechanism for concerted transcriptional actions between limb-restricted T-box factors and different members of the Hox family.

To assess the genomic target sites occupied by Tbx4 and Hoxc10 in limb buds, we performed ChIPseq for both proteins in embryonic day (E) 11.5 HL buds. We performed duplicate ChIPseq experiments for Tbx4 and Hoxc10 with two different antibodies for each protein. For Hoxc10, we only took the common peaks between two different datasets (33% and 44% of all peaks for Santa Cruz and Abcam antibodies, respectively) and isolated 1614 peaks; both antibodies revealed the same HL-specific (i.e. absent in FL) band of the expected size by western blot (Fig. S1A). For Tbx4, we kept the 4780 peaks obtained with one antibody [validated by immunohistochemistry by Ouimette et al. (2010)] because we obtained too few peaks with the other antibody, although the general gene browser profiles were similar (Fig. 1C). We analyzed the position of these peaks relative to the nearest gene and observed that both proteins predominantly lie in regions that are distant from transcription start sites (TSSs), i.e. at putative enhancers (Fig. S1B). Interestingly, 40% of Hoxc10 summits (638 sites) are located within 1 kb (the typical size of an enhancer) of the nearest Tbx4 summit (Fig. 1C) and, of these colocalized sites, 81% of the summits are located within 200 bp of each other (Fig. 1C,D). This proximity suggests that a large proportion (40%) of Hoxc10 binding targets the same putative enhancers as Tbx4, and that they bind sufficiently close to each other in the majority of cases to be part of the same enhansons (Ondek et al., 1988).

In order to define the chromatin state of Tbx4 and Hoxc10 non-promoter binding sites, we evaluated the profiles of histone marks H3K4me1 and H3K27ac, which are associated with active enhancers, as well as H3K27me3, which is associated with repressed regions (Calo and Wysocka, 2013; Ernst et al., 2011; Rada-Iglesias et al., 2011). Tbx4 and Hoxc10 both occupy genomic sites that show a bimodal enrichment of H3K4me1 and H3K27ac and a corresponding depletion of H3K27me3, suggesting that these proteins predominantly bind to active enhancers. Furthermore, sites that are co-bound by Tbx4 and Hoxc10 show higher levels of H3K4me1 and H3K27ac signal in comparison to sites bound by either Tbx4 or Hoxc10 alone (Fig. 1E, Fig. S1C). We also found through DNA sequence conservation analyses (Siepel et al., 2005) that these Tbx4-Hoxc10 co-bound regions are more highly conserved among vertebrate species than regions associated with Tbx4 or Hoxc10 binding alone (Fig. 1F,G).

To prove that Tbx4 and Hoxc10 are co-targeting enhancers within the same cells as opposed to binding common regions in different cells of the same tissue, we performed ChIP-ReChIP studies in primary E11.5 HL cells. In these cells, we see enrichment of Tbx4 at genomic regions near Gpr84 and Pck1 gene target sites chosen for their enrichment in the ChIPseq data (Fig. 2A). Sequentially, we applied Hoxc10 antibody to this Tbx4 ChIP material and further enriched for these regions, confirming that Tbx4 and Hoxc10 co-occupy common targets within the same cells (Fig. 2B). Together, these observations show that Tbx4 and Hoxc10 co-target conserved putative enhancers that have higher levels of active enhancer marks compared with putative enhancers occupied by either protein individually.

We also matched our Tbx4 and Hoxc10 peaks with published E11.5 HL CaptureC data to correlate with putative downstream target genes (Andrey et al., 2017); indeed, this work identified physical interactions between putative enhancers and 408 genes relevant to limb development. Of these 408 genes/loci, Hoxc10 targets 93 loci, while Tbx4 targets 153 loci (Fig. 2C). Of the 93 Hoxc10 target genes, 60 were co-targeted by Tbx4 and Hoxc10 at the same putative enhancer and an additional 22 loci/genes were targeted by Tbx4 and Hoxc10 individually, i.e. at different putative enhancers of the same gene. We also found that Tbx4 and Hoxc10 co-target putative enhancers of the Tbx4 and Hoxc10 genes themselves, suggesting the possibility of autoregulation through a mutual interaction platform (Table S1). One of these co-targeted genomic sites/peaks is within the Gpr84 locus illustrated in Fig. 2A, and it was identified by Andrey et al. (2017) as a putative enhancer of the Hoxc10 gene. Using the GREAT gene ontology tool (which dynamically assigns putative enhancers to target genes based on proximity), we expanded this approach and identified the putative target genes for all Tbx4 and Hoxc10 peaks. This analysis shows that 75% of Hoxc10 putative target genes are also targeted by Tbx4 and that 45% of all Hoxc10 putative target genes have an incidence of Tbx4/Hoxc10 co-binding in their vicinity (Fig. S1D). These results support a model of extensive Tbx4 and Hoxc10 collaborative function.

The finding that Tbx4 and Hoxc10 co-occupy common enhancers led us to examine whether Tbx4 and Hoxc10 interact directly. We performed a co-immunoprecipitation (Co-IP) assay using limb bud extracts. As Tbx4 is very weakly expressed in FL and Hoxc10 is not expressed (Nemec et al., 2017; Ouimette et al., 2010; Wellik and Capecchi, 2003), we used FL extracts as a negative control. In HL, we showed interaction between Tbx4 and Hoxc10 either as direct or indirect interacting partners within the same protein complex (Fig. 2D). Next, we asked whether this interaction involves DNA as a mediator. We performed Co-IP in HL bud extracts using EtBr and DNase I treatments that prevent protein-DNA complex formation (Wu et al., 2005). Hoxc10 and Tbx4 still interact in the absence of DNA (Fig. 2E). To assess whether the interaction between Tbx4 and Hoxc10 is direct or is mediated by other proteins, we carried out pull-down studies using an MBP-Tbx4 chimera together with in vitro translated Hoxc10. We found that Tbx4 and Hoxc10 interact directly (Fig. 2F).

To define the DNA-binding modalities of Tbx4 and Hoxc10 on limb enhancers, we performed de novo motif searches on the Tbx4 and Hoxc10 ChIPseq datasets. In addition to their known consensus binding sequences, we identified a novel motif containing a T-box half-site immediately adjacent to a Hox consensus sequence. Within 100 bp of the peak summits, 14% of all Tbx4 peaks (25% of the top 500) and 10% of all Hoxc10 peaks (11% of the top 500) contain this motif. This motif is also present at Tbx4 and Hoxc10 co-bound regions (17% of sites) and is further enriched in the subset of co-bound sites in which the summits of Tbx4 and Hoxc10 are no more than 100 bp apart (∼22%), suggesting Tbx4 and Hoxc10 interaction on this composite motif (Fig. 3A, Fig. S2A,B).

To determine the functional association of this interaction on the composite motif, we performed transient transfection assays with a luciferase reporter containing a trimer of the T-box–Hox composite. Transfection of Tbx4 by itself activates transcription on this reporter, whereas Hoxc10 alone has no activity. Co-transfection of Tbx4 with Hoxc10, however, revealed dose-dependent, synergistic transcriptional activation (Fig. 3B). The in vitro results show that Tbx4 and Hoxc10 interact directly and the transfection data indicate that this interaction increases transcriptional activity.

To assess the ability of Tbx4 and Hoxc10 to bind the composite motif, we performed electromobility shift assays (EMSAs) using T-box–Hox composite DNA probes. Tbx4 and Hoxc10 are both able to bind to the composite motif individually (Fig. 3C). To confirm the binding of Tbx4 and Hoxc10 to the composite, we incubated the protein extracts with Tbx4 and Hoxc10 antibodies. Whereas Tbx4 antibodies blocked Tbx4 binding, this was not seen upon incubation with Hoxc10 antibody; similarly, we did not detect Hoxc10 binding after treating extracts with Hoxc10 antibody (Fig. S2C). These results indicate that Tbx4 and Hoxc10 both have the ability to bind the T-box–Hox composite motif. We did not, however, see a super-shifted complex that could correspond to a Tbx4-Hoxc10 tri-molecular complex with the probe. This suggests that, if Tbx4 and Hoxc10 are capable of binding the composite at the same time in vivo, they might require the stabilizing presence of additional factors present in the nuclear environment that are lacking in our in vitro assay (Hellman and Fried, 2007).

To assess the requirement of Tbx4 and/or Hoxc10 binding sites for activity of the T-box–Hox composite motif, we generated three different mutants of the motif and assessed binding of Tbx4 and Hoxc10 by EMSA (Fig. 4A). When the Hox site is mutated, Tbx4 is able to bind the composite, whereas Hoxc10 cannot. When the entire T-box half-site is mutated, both Tbx4 and, surprisingly, Hoxc10 lose their ability to bind the composite. Mutating only the first 3 bp of the T-box site, however, did not prevent Hoxc10 binding, although Tbx4 binding is lost. This suggests that there is overlap between the T-box and Hox binding sites within the composite motif, as seen in Fox:Ets composite sites (De Val et al., 2008).

To assess whether the synergistic transcriptional activity observed between Tbx4 and Hoxc10 is direct or indirect, we performed luciferase assays with trimer reporters derived from the Hox mutant and 3 bp T-box mutant composite motifs (Fig. 4B,C). Tbx4 and Hoxc10 do not generate any transcriptional activity on the 3 bp T-box mutant composite; this result is expected in light of the fact that Tbx4 cannot bind this mutant and Hoxc10 has no activity on its own when acting on the intact composite motif. Interestingly, on the Hox mutant composite, the transcriptional activity of Tbx4, both alone and in combination with Hoxc10, is abolished. This result is surprising, as Tbx4 can bind this mutant motif. These observations suggest that Tbx4 transcriptional activity on the composite motif is dependent on the presence of Hox protein partners.

Enhancers are occupied by multiple TFs, often with overlapping developmental purpose, to form a single functional unit (Spitz and Furlong, 2012). Hence, we assessed whether the HL-restricted TF Pitx1 occupies similar regions as Tbx4 and Hoxc10. We observed that Pitx1 is present at 47% of Tbx4 target regions (within 1 kb), with most of these peaks (83%) within 200 bp of each other (Fig. 5A). We also found that ∼89% of Tbx4 and Hoxc10 common target regions are also occupied by Pitx1, suggesting a collaborative function between these TFs (Fig. 5B). To test if Pitx1 can interact with Tbx4 directly, we performed pull-down assays and, indeed, observed direct in vitro binding between these proteins (Fig. 5C), consistent with a model of the three HL-restricted factors acting jointly on the same set of enhancers.

Having established that Tbx4 and Hoxc10 functionally interact, we investigated whether other T-box and Hox proteins interact in a similar manner. We chose Tbx5 as our next candidate because of its paralogous relationship to Tbx4 (Minguillon et al., 2009). Tbx4 is expressed in HL and Tbx5 in FL, yet the loss of these genes leads to similar phenotypes in their respective limbs (Hasson et al., 2007, 2010; Minguillon et al., 2005; Naiche and Papaioannou, 2007). We first assessed whether these two T-box factors bind similar or different targets in their respective limb buds. Comparison of the Tbx4 ChIPseq dataset with a Tbx5 ChIPseq dataset performed in FL (GSE100734; Nemec et al., 2017) revealed a high degree of target site overlap: 47% of Tbx4 peaks in HL are within 1 kb of the position of a Tbx5 peak in FL (Fig. 5D).

We analyzed the chromatin landscapes that surround Tbx4 binding in HL or Tbx5 binding in FL. We found similar patterns of histone marks H3K4me1, H3K27ac and H3K27me3 at these sites: Tbx5 and Tbx4 both bind to active regions that show a bimodal profile for H3K27ac and H3K4me1 and a depletion of H3K27me3 (Fig. 5E). Tbx4 and Tbx5 also target similar DNA sequences, as revealed by a de novo motif search (Fig. S2A, Fig. S3) (Nemec et al., 2017). Notably, Tbx5 targets the T-box–Hox composite motif, suggesting the possibility of Tbx5 interaction with other Hox proteins. Further, we isolated the likely target genes for all Tbx4 and Tbx5 peaks. This analysis shows that 75% of Tbx4 target genes are also targeted by Tbx5 (Fig. 5F). Collectively, these results show that Tbx4 and Tbx5 share a target profile reflecting their evolutionary similarity and common function.

To assess the possibility of Tbx5 and Hox protein interactions, we compared our Tbx5 ChIPseq data with published Hoxa13 and Hoxd13 (Hoxa/d13) ChIPseq data (GSE81358; Sheth et al., 2016). The importance of the Hoxa/d13 genes for autopod development supports this choice; also, Hoxc10 is not expressed with Tbx5 in FL. The Tbx5 data were obtained with E10.5 FL, while the Hoxa/d13 analysis used distal E11.5 FL. Many Hoxa/d13 target sites share a common chromatin state at both of these time points (Sheth et al., 2016). We isolated the Tbx5-Hoxa/d13 overlapping regions and found that they have similar H3K27ac and H3K27me3 signals in E10.5 FL and distal E11.5 FL. We found that 18% of Tbx5 peaks have Hoxa/d13 binding within 1 kb, and 74% of these overlapping peaks are very close to each other (within 200 bp) (Fig. 6A,B). We also investigated whether Tbx5 and Hoxa/d13 target the composite motif in vivo. Individually, 17% of Tbx5 peaks and 16% of Hoxa/d13 peaks contain the T-box–Hox composite motif (Fig. 6C). When these two proteins are in close proximity to each other, with summits less than 100 bp apart, there is an enrichment of the composite motif (to 30%).

Next, we assessed the transcriptional interaction between Tbx5 and Hoxd13. Tbx5 activates the composite motif reporter, similar to Tbx4, highlighting their similarity (Fig. 6D). Hoxd13 did not have any activity on its own, much like Hoxc10. To our surprise, when we transfected Tbx5 and Hoxd13 together we observed a dose-dependent, antagonistic effect of Hoxd13 on Tbx5 transcriptional activity (Fig. 6D).

To investigate the nature of the Tbx5 and Hoxd13 interaction, we performed a pull-down assay between these two proteins and showed that they interact directly with each other (Fig. 6E). In light of the fact that Hoxa/d13 are also expressed in HL and Tbx4/Tbx5 have similar functions, we next asked whether Tbx4 and Hoxd13 also functionally interact. Through transfection studies, we observed a similar antagonistic interaction between Tbx4 and Hoxd13 (Fig. 6D). Also, we confirmed that Tbx4 and Hoxd13 directly interact using the pull-down assay (Fig. 6F). The negative transcriptional outcome associated with Tbx4 and Tbx5 interaction with Hoxd13, a contrast to the positive outcome of Tbx4 and Hoxc10 interaction, could reflect the role of Hoxa/d13 proteins in terminating the expression of 3′, proximally expressed Hox genes (Beccari et al., 2016; Sheth et al., 2016).

Taken together, these results show that the interaction between T-box and Hox proteins may be multifaceted and yield different transcriptional outcomes.

The combinatorial model of TF action for implementation of integrated developmental programs is supported by many studies in different systems (Spitz and Furlong, 2012). Limb development integrates positional information provided by TFs of the Hox family with the action of limb type-specific TFs such as Pitx1 and Tbx4/Tbx5. The present work uncovered novel interactions between Hox and limb type-specific TFs.

T-box TFs form a large and ancient protein family with important roles in various developmental processes (Papaioannou, 2014). Two T-box family TFs, T and Tpit (Tbx19), which are more closely related to each other than to other members of the family, play important roles in the development of the primitive streak and pituitary, respectively. Both bind to a palindromic T-box sequence (Budry et al., 2012; Kispert and Hermann, 1993; Lamolet et al., 2001; Papaioannou, 2014). By contrast, Tbx2 subfamily members only bind a T-box half-site and not a palindrome (Coll et al., 2002; Luna-Zurita et al., 2016; Sinha et al., 2000). In our work, we found that Tbx4 and Tbx5 mostly bind a T-box half-site but also a composite T-box–Hox DNA motif; further, we did not find enrichment of putative palindromic binding sites in the Tbx4 and Tbx5 ChIPseq datasets. Tbx5 was previously shown to bind sites adjacent to Nkx2-5 DNA motifs in embryonic heart (Luna-Zurita et al., 2016). Thus, T-box TFs that do not bind a palindromic T-box site might favor an association with other TFs, such as Hox factors. Whereas T and Tpit function as homodimers, the other T-box TFs may function as monomers or heterodimers with various partners.

Hox proteins also form heterodimeric complexes on DNA with Pbx and Meis TFs. These complexes form through interaction with composite DNA sequences, such as Hox:Meis or Hox:Pbx motifs, with additional proteins interacting in a DNA-independent manner (Jolma et al., 2015; Rivas et al., 2013; Shanmugam et al., 1999). In these complexes, different protein partners yield different functional outcomes (Amin et al., 2015; Gordon et al., 2010; Shanmugam et al., 1999). The presence of a T-box–Hox composite, as well as the transcriptional interactions that we observed between Tbx4/Tbx5 and different Hox proteins, suggests a framework of combinatorial interactions that is similar to Hox/Meis and Hox/Pbx.

We have shown that Tbx4 and Hoxc10 interact and co-occupy a large fraction of their genomic targets and, additionally, that they share a large fraction of putative target genes, implying extensive functional overlap. These TFs co-occupy highly conserved and active putative enhancers in vivo, consistent with the purported relevance of their synergistic transcriptional interaction. The observation that the transcriptional activity of Tbx4 depends on the presence of an intact Hox binding site, even though Tbx4 binding does not, suggests that Hox proteins are required to modulate Tbx4-dependent transcriptional activity.

The additional presence of Pitx1 at almost all of the common genomic targets of Tbx4 and Hoxc10, in conjunction with its direct interaction with Tbx4, suggests there is extensive collaboration between the HL type-specific TFs Pitx1, Tbx4 and Hoxc10 in development. This joint action might provide robustness to the HL developmental program; the combinatorial interaction of limb type-specific TFs may add a layer of regulatory complexity to allow for fine-tuning of the HL program, which depends on modulation of a highly conserved core limb development program.

We previously showed that the bulk of putative enhancers targeted by the limb-restricted TFs Pitx1 in HL and Tbx5 in FL exhibit an active chromatin status in both limbs (Nemec et al., 2017). Similarly, we now show that Tbx4 in HL and Tbx5 in FL target putative enhancers in an active chromatin state. The two limb-restricted T-box factors share about half of their putative enhancer targets, with the remaining genomic targets restricted to one T-box factor in one limb. Although Tbx4 and Tbx5 may have opposing transcriptional activities depending on the context (Ouimette et al., 2010), their interaction with other DNA-binding TFs may also contribute to limb-specific actions. The investigation of other DNA binding motifs associated with subsets of limb active enhancers binding either Tbx4 or Tbx5 revealed a frequent association with Hox and Meis target sequences. Since Meis factors often act in conjunction with Hox factors, this association may be expected.

The fact that Tbx4 interaction with Hoxd13 produces the opposite transcriptional effect as its interaction with Hoxc10 suggests that the interactions between T-box and Hox TFs could take as many forms as there are protein family members, and that these diverse interactions may lead to a wide variety of transcriptional outcomes. Specifically, the interaction between Tbx4/Tbx5 and Hoxd13 could play a role in anterior digit formation. Each of these genes individually plays a crucial role in the development of the anterior autopod (Koshiba-Takeuchi et al., 2006; Montavon et al., 2008). In mice, loss of Tbx4 in HL leads to anterior digit fusions, while loss of Tbx5 in FL leads to an extension of digit 1 and a triphalangeal thumb phenotype (Hasson et al., 2007; Naiche and Papaioannou, 2007).

The divergent transcriptional outcomes that result from Tbx4 association with Hoxc10 and Hoxd13 might well reflect domain-specific contributions: Tbx4 and Hoxc10 might be required to boost transcription of particular target genes in the proximal limb in a manner that is not needed in the distal limb, in agreement with the relative decrease in activity seen when Hoxd13 is present with Tbx4 or Tbx5. This contrasting transcriptional outcome could even be a mechanism by which Hoxa/d13 antagonize and downregulate the early and more proximally expressed Hox genes (Beccari et al., 2016; Sheth et al., 2016).

It is noteworthy that expression patterns of the limb-specific Tbx4 and Tbx5 intersect with the highly conserved and segmented expression of Hox genes in developing limb buds. Differential interactions between these TFs create the potential for diverse transcriptional outcomes. This modus operandi offers a flexible mechanism for evolutionary changes despite reliance on a limited number of limb TFs. The addition of another evolutionarily conserved HL-restricted factor, Pitx1, further increases the number of combinations that may modulate HL-specific development.

In summary, the present work proposes a model in which the combinatorial interaction between T-box and Hox TFs provides a means of compartmentalizing the functional role of these proteins in the process of limb development. The importance of specific composite target sites/enhancers could be assessed by mutagenesis, and the importance of direct Tbx-Hox interactions also assessed by mutagenesis of interaction domains once these have been identified. Protein chimeras could help define unique properties of individual Tbx or Hox partners. Although simple to list, these experiments will require significant work to figure out how a limited set of transcriptional regulators has such a substantial impact on the development of limb structures.

Chromatin immunoprecipitation (ChIP) was performed using E11.5 HL bud tissue collected from timed breeding of CD1 mice. Tbx4 and Hoxc10 ChIP was performed as described previously (Nemec et al., 2017; Sheth et al., 2016) using the following antibodies: anti-Hoxc10 (Santa Cruz, sc-33003, 7 μg per ChIP), anti-Hoxc10 (Abcam, ab153904, 7 μg per ChIP) and a homemade anti-Tbx4 (Ab1695a, 10 μg per ChIP) raised in rabbits using MBP-tagged Tbx4 epitope (aa 329-398) as published previously (Ouimette et al., 2010). The specificity of the two Hoxc10 antibodies is compared in Fig. S1A. Sequential ChIP (ReChIP) studies were performed on primary cell cultures derived from E11.5 HL buds dissociated in 0.25% trypsin, 0.1% EDTA. After 30 min, DMEM was used to quench trypsin and dissociated cells were plated on 10 cm Petri dishes. Confluent primary cells were not passaged but directly used for ChIP experiments. The first ChIP was performed with anti-Tbx4 antibody and, after the final wash, beads were treated and ReChIP was performed with anti-Hoxc10 (Abcam, ab153904) or anti-rabbit IgG antibody, as in Kim et al. (2014).

ChIPseq analysis for Tbx4 and Hoxc10 was performed as described previously (Nemec et al., 2017). For Hoxc10, only the common peaks (maximum summit distance 200 bp) obtained from two different antibodies were considered for further analysis; these represented 1614 peaks of a total of 4580 (Santa Cruz antibody) and 3629 (Abcam antibody). H3K27ac, H3K4me1, H3K27me3, Tbx5 and Hoxa/d13 ChIP datasets were retrieved from published sources (Nemec et al., 2017; Sheth et al., 2016). Hoxa/d13 peaks were isolated from common Hoxa13 and Hoxd13 (maximum summit distance 50 bp) peaks. Genome-wide correlations, heat maps and fill plots were created using EASeq (Lerdrup et al., 2016). Motif searches were performed in 200 bp windows surrounding ChIPseq peak summits using Homer (Heinz et al., 2010). For ChIP-ReChIP, relative enrichments were determined by quantitative PCR using a ViiA7-96 real-time PCR machine with SYBR Green PCR Master Mix (Thermo Fisher Scientific) and primers (Table S2). Relative enrichment is calculated as the fold-change between treatment and control conditions, each of which is normalized relative to a negative control in the Pomc promoter region. We repeated experiments twice and show representative data.

To estimate DNA sequence conservation, we used PhastCons scores downloaded from the UCSC genome browser (Siepel et al., 2005). PhastCons scores range from 0.0 (low conservation) to 1.0 (high conservation) and are calculated from the genome alignment of 60 vertebrate species. PhastCons scores were averaged at 20 bp windows across 1 kb regions flanking the peak summits. These averages were plotted directly without smoothing.

CaptureC datasets were obtained from Andrey et al. (2017). Genomic coordinates were converted from mm9 to mm10 using the UCSC liftOver tool from UCSC (http://genome.ucsc.edu/cgi-bin/hgLiftOver). ChIP summits were overlapped with CaptureC summits as reported in Andrey et al. (2017). Associations between ChIPseq peaks and putative target genes were computed with GREAT using the parameters basal-plus-extension and mouse: NCBI build 38 (McLean et al., 2010).

Protein A and Protein G Dynabeads (Invitrogen) were incubated overnight with anti-Hoxc10 (Santa Cruz, sc-33003, 8 μg) or anti-goat IgG antibodies (Santa Cruz, sc-2028, 8 μg). Limb buds collected from E11.5 mouse embryos (CD1) were washed with PBS and mechanically homogenized. Cell extracts and Co-IP incubations were performed as described previously (Budry et al., 2011). PVDF membranes (Millipore) were blotted with a homemade anti-Tbx4 antibody (1/1000) raised in rabbits using 100 μg MBP-tagged Tbx4 epitope (aa 1-72) and anti-Hoxc10 antibody (Santa Cruz, sc-33003, 1/200). Where appropriate, limb bud extracts were incubated with DNase I (Invitrogen) or ethidium bromide (EtBr) as described previously (Wu et al., 2005).

MBP fusion proteins coupled to maltose-labeled agarose beads were produced as described (Bilodeau et al., 2006). ³⁵S-labeled proteins were synthesized in vitro using a TNT Coupled Reticulocyte Lysate System Kit (Promega). Labeled proteins were incubated with MBP-tagged proteins in 200 μl TNEN₅₀ (50 mM Tris pH 7.5, 5 mM EDTA, 50 mM NaCl, 0.1% NP-40) with 1 mM PMSF and 2% BSA for 4 h at 4°C. Beads were washed three times with 1 ml TNEN₁₂₅. Bound proteins were resolved by SDS-PAGE and visualized by autoradiography.

HEK 293T cells were transfected using the calcium phosphate co-precipitation method: 400,000 cells per well were plated in 12-well plates and transfected with 1 μg total DNA (50 ng reporter, 500 ng effector or empty vector, 50 ng pRL-CMV as internal control, and adjusted to 1 µg total with pBluescript SK+). Medium was changed after 24 h and cells were harvested 36 h after transfection. Firefly luciferase and Renilla luciferase activities were assayed in lysates with a luminometer using the luciferin (Becton Dickinson) and coelenterazine (NanoLight Technology) substrates, respectively. All experiments were performed using three biological and three technical replicates. Statistical analysis was performed using Student's t-test with the Benjamini-Hochberg correction (Benjamini, 1995).

To obtain relevant proteins, expression plasmids were transfected in HEK 293T cells in a 10 cm Petri dish. After 36 h, cells were washed with PBS and collected. Cells were first washed and then resuspended with 400 μl cold Buffer A (10 mM HEPES pH 7.9, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM DTT, 1 mM PMSF, 10 μg/ml aprotinin, 1 μg/ml pepstatin, 1 μg/ml leupeptin) and incubated for 15 min on ice. Cells were further treated with 50 μl NP-40 (10%) and vortexed for 10 s. Supernatant was removed and 50 μl Buffer C (20 mM HEPES pH 7.9, 400 mM KCl, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 1 mM PMSF, 10 μg/ml aprotinin, 1 μg/ml pepstatin, 1 μg/ml leupeptin) was added to the pellet. Pellets were kept on mild vortex for 30 min at 4°C and then centrifuged at 14,000 rpm (12,000 g) for 5 min at 4°C. Supernatant was collected and proteins quantitated using the Bradford assay. For binding assays, 20 μg protein was added to a 19 μl volume containing 3 μg dA-dT plus dI-dC and 5 μl of 4× buffer (100 mM HEPES pH 7.9, 335 mM KCl, 40% glycerol, 20 mM DTT) and incubated for 15 min. Probes were prepared as described (Maira et al., 1999) using Micro Bio-Spin Bio-Gel P-30 chromatography columns (Bio-Rad). After 15 min, 1 μl probe (50,000 cpm) was added to the reaction mixture and incubated for 1 h at 4°C prior to being loaded onto gels as described (Maira et al., 1999). Antibodies were incubated with the nuclear extracts for 30 min on ice after addition of probes.

Reporter plasmids were constructed by cloning three copies of wild-type or mutant T-box–Hox composite fragments (see Table S2) upstream of a short RSV promoter (extending as far upstream as −130 bp) in the PGL4.10 luciferase vector (Promega). Effector plasmids were constructed by cloning Tbx4 (NM_011536), Tbx5 (NM_011537), Hoxc10 (NM_010462) and Hoxd13 (NM_008275) cDNAs in the pRSV-Luc vector described above.

We thank Isabelle Brisson, Dimitar Dimitrov and Sara Demontigny for their work in the IRCM Animal Care Facility and Évelyne Joyal for secretarial assistance.