Bmp signaling regulates a dose-dependent transcriptional program to control facial skeletal development

The CNC is a migratory cell population that originates in the dorsal neural tube and diversifies into multiple cells types, including cartilage, bone, neurons and glia. Much of the craniofacial skeleton such as the skull vault or calvarium, mandible and midface develops through direct, intramembranous ossification of CNC-derived progenitor cells (Chai and Maxson, 2006). For example, in the skull, osteogenesis occurs in discrete areas within the cranial mesenchyme, resulting in the flat bones of the skull forming between the central nervous system (CNS) and overlying ectoderm. The mandible and most midfacial bones develop by direct ossification of CNC-derived branchial arch mesenchyme.

Bmp signaling plays crucial roles in normal craniofacial development, and Bmp4 deficiency results in craniofacial anomalies, such as cleft lip and palate, in mouse and humans (Liu et al., 2005b; Suzuki et al., 2009). In the mandible, Bmp4 regulates proximodistal patterning and timing of bone differentiation in mandibular mesenchyme (Liu et al., 2005a; Merrill et al., 2008). There is strong evidence that Bmp signaling regulates craniofacial morphological change during evolution. Both gain-of-function studies and comparative expression data revealed Bmp4 to be a crucial regulator of beak shape in Darwin’s finches, a classic model of evolutionary diversification (Abzhanov et al., 2004; Wu et al., 2004). Other experiments in Cichlid fish also support the notion that Bmp4 is a major regulator of craniofacial cartilage shape and morphological adaptive radiation (Albertson et al., 2005). However, the downstream genes that Bmp regulates to control facial morphogenesis are poorly understood.

Recent molecular insights into Bmp signaling indicate that the downstream effector mechanisms for Bmp signaling are complex and require further study (Wang et al., 2011). The canonical Bmp pathway involves nucleo-cytoplasmic shuttling of Smad effectors in response to Bmp signaling. In addition, Smad-independent mechanisms that are mediated through MapK pathways are also known to play an important role in tooth development (Xu et al., 2008). More recent work, uncovering a third Bmp effector mechanism, revealed that Smad1/5 directly binds to the Drosha complex to promote microRNA (miR) processing (Davis et al., 2008). In addition, Bmp signaling can induce miR transcription through a canonical Smad-regulated mechanism (Wang et al., 2010). Despite the central importance of Bmp signaling for craniofacial development, congenital defects, and evolution, the mechanisms underlying Bmp action in CNC remains poorly understood.

Here, we have investigated Bmp signaling in CNC development using both gain- and loss-of-function approaches. Inactivation of Bmp2, Bmp4 and Bmp7 in CNC indicate that Bmp2 and Bmp4 are the major Bmp ligands required for development of CNC-derived bone and cartilage. Moreover, gain-of-function studies indicate that elevated Bmp4 in CNC results in dramatic changes in the facial skeleton. Expression profiling and quantitative RT-PCR (qRT-PCR) studies uncovered a common set of Bmp regulated target genes in both gain- and loss-of-function embryos. A subset of Bmp-regulated targets were directly bound by Smad 1/5, indicating direct regulation. Bmp-regulated genes control self-renewal, osteoblast differentiation and negative feedback regulation, suggesting that Bmp signaling regulates facial skeletal morphogenesis by controlling the balance between self-renewing progenitors and differentiating lineage-restricted cells.

The generation and characterization of Bmp2 and Bmp4 conditional null mice and Wnt1Cre transgene mice has been previously described (Chai et al., 2000; Danielian et al., 1998; Liu et al., 2004; Ma and Martin, 2005). The conditional Bmp7-null allele will be described elsewhere (Y.B. and J.F.M., unpublished).

Embryos were placed overnight in water and scalded in hot water for 30 seconds. The skin and internal organs were removed and the sample fixed overnight in 95% ethanol. The cartilage was stained with 0.15 mg/ml Alcian Blue (Sigma) in a 1:4 mixture of glacial acetic acid and 95% ethanol. After staining overnight, the samples were rinsed twice in 95% ethanol and incubated for 24 hours in 95% ethanol. In preparation for bone staining, the sample was placed in 2% KOH for 1 hour. Bones were stained using 0.05 mg/ml of Alizarin Red (Sigma) diluted in 2% KOH for 2-4 hours and then cleared with glycerol. For adult skulls, partially dissected heads were placed with Dermestid Beetles (Carolina) to remove soft tissues. After soft tissue removal, skulls were placed in 100% ethanol to remove any remaining beetles, air dried and stored at room temperature.

E14.5 embryos were dissected, decapitated and fixed in 4% paraformaldehyde for 1 hour. Heads were bisected midsagittally and the brain and dura mater removed, washed three times in PBS and NTMT. The NBT/BCIP (Roche) substrate in NTMT was added until bone staining was observed. Samples were then washed in PBS, and post-fixed in 4% paraformaldehyde.

Cell lysates were prepared by dissecting E11.5 mandibles in PBS. Tissue was homogenized in lysis buffer (50 mmol/l Tris, 150 mmol/l NaCl, 1% Triton, 0.5% deoxycholate plus protease inhibitors cocktail; Roche, EDTA and sodium vanadate; Sigma). Protein concentration was determined by the Protein Assay Reagent kit (Pierce). Whole-cell lysates were separated by 12% SDS-PAGE and transferred to nitrocellulose. After blocking with 5% nonfat milk at room temperature for 1 hour, blots were incubated with the p-Smad 1/5/8 polyclonal antibody (1:1000 dilution; Cell Signaling, #9511) and Smad 1/5/8 polyclonal antibody (1:1000 dilution; Santa Cruz, sc-6031 X) at room temperature with agitation for 1 hour, followed by incubation with anti-rabbit IgG horseradish peroxidase-conjugated antibody (1:2000; GE Healthcare). Blots were developed using an enhanced chemiluminescence detection kit (ECL; Pierce). To confirm equal loading, we used anti-β-actin antibody (1:5000; Sigma). Western blots were quantified by densitometry analysis using Image J software (NIH).

Hematoxylin-Eosin staining and whole-mount in situ hybridization were performed as previously described (Wang et al., 2010). Plasmids for in situ probes have been previously described: Dlx6 (Charite et al., 2001), Gata3 (Ruest et al., 2004), Hand1 (McFadden et al., 2005), Smad6 (Ma et al., 2005), Msx1 (Ishii et al., 2005), Msx2 (Ishii et al., 2003) and Tbx20 (Shen et al., 2011). Full-length cDNA for mouse BMP4 was provided by Dr Stephen Harris (UTHSC, San Antonio, TX, USA) and was linearized with SpeI and transcribe with T7. Exons 3 and 4, including 3′UTR of Gadd45g were amplified and subcloned into T-easy vector. Plasmid was linearized with EcoRV and transcribed with T7 RNA polymerase. Cux2 (exon 23) and Satb2 (exon 10) were amplified and subcloned into T-easy vector. Plasmid was linearized with SacI and transcribed with T7 RNA polymerase. For all experiments, at least three mutants and three controls embryos were analyzed for each probe.

To generate the Bmp4^tetO allele, we constructed a targeting vector that resulted in a 665 bp deletion upstream of and including the Bmp4 basal promoter and Bmp4 exon 1. We modified the tetO plasmid, a kind gift from Raymond MacDonald’s laboratory (UT Southwestern Medical Center, TX, USA), which contains the tetracycline operator (tetO7), CMV promoter, IRES-lacZ and a poly-adenylation sequence. Bmp4 genomic DNA was isolated from the 129/S BAC library. A 9 kb EcoRI fragment of Bmp4 genomic DNA was subcloned into pBluescript (Stratagene). We inserted Bmp4 cDNA into NotI and KasI sites downstream of the tetO7 and CMV promoter. The phosphoglycerol kinase neomycin-resistance cassette (pgk-neo) with two flanking Frt sites was blunt cloned into the AscI site downstream of the IRES-lacZ. The 3 kb 3′ homology arm was amplified by PCR using the primers (5′ to 3′) tgagcagggcaacctggagaggg and tccgaatggcactacggaatggct, and blunt-end ligated into a SwaI site downstream of pgk-neo. The Bmp4^tetO 5′ arm, 1.7 kb EcoRI-ApaLI fragment, was cut from Bmp4 genomic DNA and blunt end cloned into XhoI site upstream of the tetO7. The targeting construct was linearized with PmeI and electroporated into AK7 embryonic stem cells, selected in G418, and screened by Southern blot. The cells were digested with EcoRV and Bmp4 exon 4 was used as the 3′ flanking probe. Wild-type allele gives a 22 kb EcoRV fragment and the targeted allele gives a 16 kb EcoRV fragment. The targeting frequency for the tetO allele is 4.2% (7/167). We also used a SpeI digest to confirm correct targeting: the wild-type band was 6.4 kb, the cDNA band was 11.7 kb and the mutant allele was 8.0 kb. We used the Rosa 26 rtTA allele (Belteki et al., 2005) to express the reverse tetracycline activator (Tet-on) under the control of the Wnt1Cre allele. This will drive Bmp4^tetO allele expression in neural crest cells in the presence of doxycycline.

Pregnant females were given doxycycline (Sigma) in the drinking water (2 mg/ml) and in the food (Bio-Serv; 200 mg/kg), for a 24 hour period unless otherwise specified.

E11.5 mandibles were dissected in ice-cold PBS and placed in RNAlater (Ambion) for RNA stabilization. mRNA was then extracted using the RNeasy Micro Kit (Qiagen). First strand cDNA synthesis was then performed utilizing the SuperScript II Reverse Transcriptase kit (Invitrogen) with 500 ng of mRNA. Quantitative real-time PCR was performed using SYBR Green QPCR Master Mix (Invitrogen) in triplicate reactions and ran on Mx3000P thermal cycler (Stratagene). Primers used in this study are listed in supplementary material Table S2. For all qRT-PCR experiments, at least three mutants and three controls embryos were analyzed. DNA microarray analysis, including gene ontology analysis, was performed using the OneArray Mouse Whole Genome Array (Phalanx Biotech Group). Mandibular processes were pooled to collect a minimally required RNA amount: seven Bmp4 OE and five controls embryos were used. Gene ontology results were confirmed using DAVID Bioinformatics Resources 6.7 National Institute of Allergy and Infectious Diseases (NIAID), NIH. Microarray data have been submitted to Gene Expression Omnibus (Accession Number GSE35402).

Chromatin immunoprecipitation (ChIP) analysis was performed as previously described (Wang et al., 2010). We used E11.5 wild-type mice mandibles and the mouse osteoblastic cell line MC3T3-E1 (ATCC). Cells where maintained and propagated following ATCC protocols. Experiments were performed with 90% confluent cultures and Bmp4 (R&D System) was added to the media for a final concentration of 25 pg/μl for 12 hours. Primer sequences used for amplification of the Bmp/Smad regulatory elements are found in supplementary material Table S2.

For sequence analysis and multiple sequence alignment, Ensembl genome database and MAFFT (http://mafft.cbrc.jp/alignment/server/) were used.

We used the Wnt1Cre driver and a Bmp4 conditional null allele, Bmp4^floxneo, to inactivate Bmp4 in CNC (Chai et al., 2000; Liu et al., 2004). Intercrosses between Wnt1Cre; Bmp4^floxneo/+ with Bmp4^floxneo homozygous mice generated Wnt1Cre; Bmp4 ^{floxneo;floxneo(f/f)} mutant embryos, hereafter called Bmp4 CKO. Evaluation of skull preparations indicated that Bmp4 CKO mutant newborn skulls had enlarged frontal fontanelle and subtle mandibular defects (Fig. 1A,B). Examination of Msx1 and Msx2 expression, markers of preosteogenic head mesenchyme, revealed that these markers were reduced but still present in the Bmp4 CKO embryos (supplementary material Fig. S1A-D). Alkaline phosphatase, a marker of both preosteoblasts and mature osteoblasts, was also reduced in the frontal bone of Bmp4 CKO embryos, indicating a defect in the transition from pre-osteoblast to osteoblast in the Bmp4 CKO (supplementary material Fig. S1E,F). Persistent Msx gene expression in Bmp4 CKO embryos suggested that other Bmp ligands probably had overlapping functions with Bmp4 in CNC.

To test this idea, we generated compound conditional loss-of-function mutants for Bmp2, Bmp4 and Bmp7 using Bmp conditional null alleles and the Wnt1Cre driver. The genetic analysis indicated that although Bmp4 was the major functional Bmp ligand in CNC-derived bone development, Bmp2 also had important functions.

Analysis of Bmp4 and Bmp7 compound mutants indicated that Bmp7 failed to have a significant influence on Bmp4 CKO mutant phenotype (Fig. 1A-E,R). By contrast, Bmp2 deletion in the Bmp4 CKO background resulted in a significant worsening of frontal and mandibular bone phenotypes. The Bmp2/4 CKO mutant had a drastic reduction in most CNC-derived bones such as the frontal and mandible (Fig. 1F-I,S). Additionally, Bmp2 had a unique role in coronoid process development (Fig. 1G,K,N). Analysis of triple-mutant combinations further supported the conclusion that Bmp4 and Bmp2 were the major ligands in frontal and mandibular bone development. Comparison between embryos that were Bmp2; Bmp4 compound homozygous and were either Bmp7^+/+, Bmp7^flox/+ or Bmp7^flox/flox indicated that Bmp7 has a minor role in frontal bone formation (Fig. 1I,P,Q). In addition to the frontal and mandibular bone defects, other CNC derived bones were affected in Bmp compound mutants (supplementary material Fig. S2).

We next tested the role of elevated Bmp4 signaling in CNC. We developed a tetracycline-regulated Bmp4 allele (Bmp4^tetO) by replacing the first non-coding exon and basal promoter region of Bmp4 with a tet operator Bmp4 cDNA fusion gene (Fig. 2A,B). We used a cre-recombinase inducible rtTA (Tet-on) allele, R26R^rtTANagy allele and the Wnt1Cre allele to induce rtTA in the CNC lineage (Belteki et al., 2005). Bmp4 was overexpressed by inducing Wnt1Cre; Bmp4^tetO/+; R26R^rtTANagy (Bmp4 OE) embryos with doxycycline (dox) (Fig. 2C).

To determine whether Bmp4 OE embryos had elevated Bmp4 levels in CNC, we dissected mandibular processes from Bmp4 OE and control embryos and performed qRT-PCR. Bmp4 OE embryos had approximately 30-fold inducible Bmp4 upregulation. Furthermore, Bmp4 levels were increased with higher levels of dox (Fig. 2D). Bmp4 induction could be detected 3 hours after dox induction and increased through 24 hours of induction (Fig. 2E). Elevated Bmp4 was also detectable in CNC by whole-mount in situ in E11.5 embryos (Fig. 2F). Western blot indicated that the elevated Bmp4 mRNA expression resulted in approximately a 2.5-fold increase in p-Smad1/5/8 activity in the mandibular process (Fig. 2G). In addition, p-Smad1/5/8 activity in the Bmp4^tetO heterozygous embryos was reduced when compared with the wild-type control, indicating that Bmp4^tetO allele is a Bmp4 hypomorphic allele (Fig. 2G; supplementary material Fig. S3). Analysis of Bmp4^tetO homozygous mutant embryos supported the conclusion that Bmp4^tetO is hypomorphic (supplementary material Fig. S3).

We evaluated the phenotypic effect of elevated Bmp4 expression in CNC by analyzing E16.5 Bmp4 OE embryos that had been induced at embryonic stages between E10.5 and E14.5. Bmp4 induction at E13.5 resulted in a mandible that was more pointed in appearance although overall facial form was not significantly changed (Fig. 3A-D). Bmp4 induction at E11.5 resulted in a mandible that was shorter and more pointed than the control (Fig. 3E-H). In E10.5 Bmp4-induced embryos, the face was drastically changed. There was shortening and pointed appearance in both the mandible and maxilla. The overall shape of the head was more rounded when compared with the control mouse embryo. Moreover, the orientation of the eyes was more anterior when compared with the control (Fig. 3I-L).

Skeletal preparations indicated that, at E10.5, Bmp4 induction resulted in strong reduction of rostral bony elements, such as nasal bones, with a drastically shortened face (supplementary material Fig. S4G,H). Overall bone quality was defective in that skull bones revealed multiple translucent areas. Bmp4 induction at E12.5 had less dramatic morphological changes but the size of the nasal cartilages was expanded while nasal and frontal bones were absent or reduced (supplementary material Fig. S4E,F) and the mandible was shorter. Induction at E13.5 revealed reduction in nasal bones and coronoid process of the mandible (supplementary material Fig. S4C,D).

Histological analysis indicated that Bmp4 E12.5 induction resulted in a large increase in both nasal and Meckel’s cartilage, indicating that Bmp4 modulates both cartilage and bone development (supplementary material Fig. S5A-D,I-L). These Bmp4 E12.5 induced embryos also had cleft palate (supplementary material Fig. S5K). Embryos induced at E14.5 showed milder phenotypes, indicating that Bmp4-induced facial changes are stage dependent (supplementary material Fig. S5E-H).

To comprehensively investigate genes regulated by Bmp signaling in the mandibular process, we performed expression profiling using RNA extracted from E11.5 Bmp4 OE mandibles that were induced with dox for 24 hours. Using a twofold change (P<0.05) as threshold, we identified 144 downregulated and 120 upregulated genes (supplementary material Fig. S6A and Table S1). Although gene ontology analysis for all genes revealed several gene clusters involved in multiple cellular processes, among Bmp4-induced genes, gene ontology indicated that transcriptional regulation was the main cellular process influenced (supplementary material Fig. S6B,C).

Included in the induced BIG signature after qRT-PCR validation were multiple transcription factor families, including Gata genes, Hand genes, Satb genes and Klf genes (Fig. 4A,B). Other upregulated transcriptional regulators include Atf3, Cux2 and Isl1. Gata3, Hand1 and Satb2 are transcription factors that play important roles in craniofacial development (Fig. 4A,B). Importantly, many of these genes (such as Isl1 and Id1) have been shown to be targets of Smad-mediated signaling in other developmental processes and many have known roles in craniofacial development (Chai and Maxson, 2006). Interestingly, negative regulators of Bmp-Smad signaling, Gadd45 and Smad6, and Noggin, were also strongly upregulated in the Bmp4 OE mandibles, uncovering a negative-feedback pathway in the mandible. We validated the microarray results by qRT-PCR using independently generated RNA from control and Bmp4 OE mandibles (Fig. 4B). We also looked at expression of essential transcriptional regulators of cartilage and bone, Runx2 and Sox9, in Bmp4 OE mandibles. Interestingly, although Sox9 was significantly upregulated, Runx2 was mildly downregulated, suggesting that Bmp may regulate osteoblast genes both cooperatively and independently of Runx2 (supplementary material Fig. S7).

In situ hybridization demonstrated an expanded expression pattern of the BIG signature after Bmp4 induction (Fig. 4C; supplementary material Fig. S8). Interestingly, there were distinct upregulated gene expression patterns. Some genes, such as Hand1 and Smad6, were broadly expanded throughout the CNC, whereas other genes, such as Gadd45g and Satb2, were expanded in a more restricted region of the mandible (Fig. 4C).

To determine whether the BIG signature was also reduced in the Bmp loss-of-function embryos, we evaluated the expression of the candidate genes in Bmp-deficient embryos. Because the Wnt1Cre; Bmp2/4/7 triple mutants are recovered at low frequencies, we supplemented our experiments with the Nkx2.5Cre; Bmp4^n/f embryos that have greatly reduced Bmp signaling in the mandibular ectoderm and mesenchyme (Liu et al., 2005a). qRT-PCR analysis of the mandibular tissue showed significant reduction in the expression levels of genes that were upregulated in the Bmp4 OE tissue (Fig. 5A). Moreover, whole-mount in situ analysis for a subset of these genes revealed loss of mandibular expression in Bmp4 mutant embryos (Fig. 5B). qRT-PCR data from Wnt1Cre; Bmp2/4/7 triple mutant mandibles was consistent with the data from the Nkx2.5Cre; Bmp4^n/f embryos (data not shown).

To determine whether genes in the BIG signature were directly regulated by Smad-mediated transcription, we undertook a bioinformatic approach to identify conserved Smad recognition elements within a 5 kb region located in the 5′ flanking region of these genes. For Gadd45g, Gata3, Hand1, Satb2 and Smad6, we uncovered phylogenetically conserved Smad recognition elements (Fig. 6A). We tested the ability of Smad1/5 to bind to these sequences in vivo by ChIP assays in wild-type mandibles and in the mouse osteoblastic cell line MC3T3-E1. Our data indicate that in the developing mandible, Smad1/5 binds directly to the chromatin of these five genes (Fig. 6B). Moreover, in MC3T3-E1 cells that were cultured in the presence of Bmp4, we found an enrichment in Smad1/5 chromatin binding after Bmp treatment (Fig. 6C).

We performed a comprehensive analysis of Bmp function in CNC using genetics, gene expression profiling and ChIP. Our profiling and qRT-PCR validation data from the Bmp4 OE embryos indicate that the BIG signature contains 21 genes. Moreover, we validated 17 of the 21 genes in the Bmp loss-of-function model.

Within the BIG signature are transcriptional regulators important for osteoblast differentiation and progenitor cell self-renewal. Bmp signaling also induces a negative regulatory pathway that probably functions as a buffering mechanism to maintain precise Bmp signaling levels. We show that five genes, Gata3, Gadd45, Hand1, Satb2 and Smad6, are directly bound by Smad1/5. Our findings suggest that a balance between self-renewal and progenitor differentiation in CNC underlies Bmp-regulated facial skeletal development (Fig. 6D).

Our data show that Bmp signaling in CNC progenitors regulates genes that are directly implicated in self-renewal such as Id and Klf genes. Klf2 and Klf5, as well as Id1 and Id4 are regulated by Bmp signaling in CNC progenitors. Klf genes, regulators of ES cell self-renewal through the control of Nanog expression and cellular reprogramming, have not been shown to be regulated by Bmp signaling. Our findings suggest that Bmp signaling in CNC promotes self-renewal of CNC cells that allow progenitor cells to persist as craniofacial development progresses.

In other in vivo models systems such as the Drosophila ovary, Bmp signaling promotes self-renewal and proliferation of somatic stem cells and prolongs progenitor lifespan (Kirilly et al., 2005). In CNC progenitors, it is conceivable that Bmp signaling increases the number of self renewing progenitor cells in addition to activating expression of the CNC self-renewal program.

In addition to inducing Id1 expression in embryonic stem cells, Bmp signaling directly promotes self-renewal in collaboration with leukemia inhibitory factor through a direct interaction between Smad1 and the core self-renewal factors Nanog, Oct4 and Sox2 (Chen et al., 2008; Fei et al., 2010; Ying et al., 2003). As defined by ChIP seq, Smad1 commonly occupies Nanog-Oct4-Sox2 bound loci, revealing that Bmp signaling directly interacts with the core pluripotency machinery to enhance pluripotency and self-renewal (Chen et al., 2008). Moreover, the Smad1-containing complexes in ES cells recruit the HAT p300 to activate gene transcription. Our findings support a model in which Bmp signaling enhances CNC progenitor self-renewal by activating the self-renewal gene program (Fig. 6D). Future ChIP seq experiments will be required to determine whether Smad1 directly interacts with a pluripotency program in CNC progenitors.

Bmp signaling also regulates lineage-restricted genes such as Satb2 that enhance osteoblast lineage development. Satb2 is a DNA-binding and architectural factor that has a positive role in osteoblast development. Satb2 deficiency results in phenotypes that are similar to mild Bmp loss-of-function phenotypes such as cleft palate and calvarial defects with shortened mandible. Importantly, Satb2 controls osteoblast differentiation through regulation of Runx2 and Atf4 expression (Dobreva et al., 2006). Notably, similar to Bmp4, Satb2 deficiency causes orofacial clefting in humans as well as mice (Britanova et al., 2006). Our data showing that Satb2 is a direct Smad1/5 target indicate that a major pathway for Bmp-regulated facial bone development is through Satb2 function (Fig. 6D).

Similar to Bmp4 loss-of-function mutants, Gata3 mutants have a medial mandibular deficiency (Liu et al., 2005a; Ruest et al., 2004). There is evidence that Gata3 directly regulates Nmyc in the branchial arches, suggesting that one cellular mechanism in the Gata3 mutant mandible is reduced proliferation. Other data also indicate that Gata3 promotes osteoblast and neuron survival, suggesting that, in addition to proliferation, apoptosis may also be enhanced in Gata3 mutants as it is in Bmp4 loss-of-function embryos (Chen et al., 2010; Liu et al., 2005a; Tsarovina et al., 2010).

Hand1 and Hand2 have overlapping function in medial mandible development, and promote progenitor cell proliferation and inhibit differentiation (Barbosa et al., 2007; Funato et al., 2009). In the heart, Hand1 overexpression increases cell proliferation, indicating that the Hand mandibular defect may result from reduced progenitor cell proliferation. It is interesting to note that both Hand1 and Gata3 have been implicated in trophoblast development, indicating that these two genes may have interrelated functions in multiple cellular contexts (Ralston et al., 2010).

In addition to Bmp signaling, the endothelin (Edn) signaling pathway regulates Hand gene expression and also is crucially important in facial form regulation. In zebrafish, an interplay between Bmp and Edn signaling is important for branchial arch dorsoventral patterning (Alexander et al., 2011; Zuniga et al., 2011). Edn-deficient embryos display mandibular-to-maxillary transformations that are restricted to the Hox-negative CNC (Gitton et al., 2010). Moreover, gain-of-function experiments indicate that expanded Edn and Hand2 in maxillary process results in transformation to a mandibular phenotype. Our data indicate that Bmp and Edn signaling converge on Hand gene function to regulate facial development.

The Dlx5/6 genes, targets for Edn signaling and direct regulators of Hand2, are also crucially important in facial development (Depew et al., 2002). Only modest Dlx6 expansion in Bmp4 OE embryos indicates that Bmp induced regulation of Hand genes primarily goes directly through Smad1/5 (supplementary material Fig. S8).

Negative autoregulation is a mechanism to confer robustness to the developing embryo by buffering the system from elevated Bmp levels and is crucial for normal craniofacial and heart development (Paulsen et al., 2011; Prall et al., 2007). The negative regulatory loop also probably has an impact on the severity of craniofacial phenotypes that we observed. For example, in Wnt1Cre; Bmp7 CKO embryos, our preliminary data indicate that Bmp2, but not Bmp4, expression is upregulated, suggesting that Bmp7 inhibits Bmp2 expression through an unknown mechanism. This autoregulation may account for the relatively mild phenotypes seen in Bmp7 mutant embryos.

Bmp4-induced expression of negative regulators of Bmp signaling such as Smad6, Noggin and Gadd45g (Fig. 6D). The importance of finely tuned Bmp signaling levels in mice and humans is apparent from Noggin loss-of-function studies in mice, as well as human genetics studies. Noggin deficiency results in cleft palate, defective mandibular development, as well as limb and heart defects (Brunet et al., 1998; Choi et al., 2007; Gong et al., 1999; He et al., 2010). The negative auto-regulatory pathway includes two genes, Gadd45g and Smad6, that are Smad-regulated direct Bmp targets. Moreover, the induced negative regulatory genes modulate the pathway by multiple mechanisms. Smad6 inhibits R-Smad activity by both competing for Smad4 and also inhibiting R-Smad phosphorylation. Gadd45g promotes ubiquitin ligase interaction with the Smad linker region to destabilize R-Smad, while Noggin is a competitive inhibitor of the Bmp ligand-receptor interaction (Sheng et al., 2010).

Cux2 has not been previously connected to Bmp signaling or bone morphogenesis. The closely related gene Cux1 is regulated by Tgfβ and represses collagen expression (Fragiadaki et al., 2011). Cux2 is regulated by Notch signaling in the spinal cord, where it controls the balance between neural progenitors and differentiated neurons by modulating cell cycle progression and enhancing interneuron fate (Iulianella et al., 2008; Iulianella et al., 2009). Interestingly, Notch has been shown to work in parallel with Bmp signaling in valve development, suggesting that Notch and Bmp signaling may work together to regulate the balance of CNC progenitors with differentiated skeletal cells (Luna-Zurita et al., 2010).

The BIG signature contains genes that are known to be Bmp regulated. Previous experiments uncovered a Bmp-Msx genetic pathway in multiple contexts within the developing craniofacial apparatus, including skull, palate and teeth (Chai and Maxson, 2006). Our data also indicate that Isl1is regulated by Bmp signaling during mandibular development. There is evidence that Bmp and Isl1 function in a positive-feedback loop in the mandible (Mitsiadis et al., 2003). Our data support these earlier findings and substantially extend previous understanding of Bmp targets in craniofacial development.

Our in situ data indicate that in the Bmp4 OE embryos one gene group, including Gata3, Hand1 and Smad6 was broadly expressed throughout the cranial neural crest in response to Bmp4 overexpression. By contrast, genes such as Satb2 and Gadd45g were more resistant to Bmp4 overexpression and were upregulated in a more discrete spatial pattern. This observation indicates that other regulatory mechanisms, such as microRNA (miR)-mediated regulation, fine-tune the expression of Satb2 and Gadd45g. miRs are small non-coding RNAs that post transcriptionally regulate gene expression by enhancing mRNA degradation or by inhibiting translation (Bartel, 2009). One idea is that other signaling pathways, such as Notch or Wnt, may regulate miR expression that then degrades target mRNAs in specific domains of the branchial arches. Alternatively, the distinct responses to Bmp4 signaling may be regulated at the level of chromatin regulation such that Smad-mediated activation is offset by negative regulatory mechanisms that cannot be overcome by high levels of Bmp signaling. Future experiments are required to investigate these ideas.

We thank A. Bradley, and R. Johnson for reagents, and James Douglas Engel, Richard Maas, Robert E. Maxson, Jr, Sylvia Evans and Stephen E. Harris for in situ probes. Part of this work was included within the PhD theses of J.S. and M.B.-C.