TGF-β-activated Kinase 1 (Tak1) Mediates Agonist-induced Smad Activation and Linker Region Phosphorylation in Embryonic Craniofacial Neural Crest-derived Cells*

Background: The role of Smad-independent TGF-β signaling in craniofacial development is poorly elucidated. Results: In craniofacial mesenchymal cells, Tak1 regulates both R-Smad C-terminal and linker region phosphorylation in TGF-β signaling. Conclusion: Tak1 plays an irreplaceable role in craniofacial ecto-mesenchyme during embryogenesis. Significance: Understanding the mechanisms of TGF-β signaling contributes to knowledge of pathogenetic mechanisms underlying common craniofacial birth defects. Although the importance of TGF-β superfamily signaling in craniofacial growth and patterning is well established, the precise details of its signaling mechanisms are still poorly understood. This is in part because of the concentration of studies on the role of the Smad-dependent (so-called “canonical”) signaling pathways relative to the Smad-independent ones in many biological processes. Here, we have addressed the role of TGF-β-activated kinase 1 (Tak1, Map3k7), one of the key mediators of Smad-independent (noncanonical) TGF-β superfamily signaling in craniofacial development, by deleting Tak1 specifically in the neural crest lineage. Tak1-deficient mutants display a round skull, hypoplastic maxilla and mandible, and cleft palate resulting from a failure of palatal shelves to appropriately elevate and fuse. Our studies show that in neural crest-derived craniofacial ecto-mesenchymal cells, Tak1 is not only required for TGF-β- and bone morphogenetic protein-induced p38 Mapk activation but also plays a role in agonist-induced C-terminal and linker region phosphorylation of the receptor-mediated R-Smads. Specifically, we demonstrate that the agonist-induced linker region phosphorylation of Smad2 at Thr-220, which has been shown to be critical for full transcriptional activity of Smad2, is dependent on Tak1 activity and that in palatal mesenchymal cells TGFβRI and Tak1 kinases mediate both overlapping and distinct TGF-β2-induced transcriptional responses. To summarize, our results suggest that in neural crest-derived ecto-mesenchymal cells, Tak1 provides a critical point of intersection in a complex dialogue between the canonical and noncanonical arms of TGF-β superfamily signaling required for normal craniofacial development.

Members of the TGF-␤ superfamily are involved in many normal and pathologic events including development, inflammation, fibrosis, and cancer (1). TGF-␤s and bone morphogenetic proteins (BMPs) 3 form two important subgroups within the superfamily (2,3). They bind to largely subfamily-specific membrane-spanning complexes of Type I and Type 2 receptors resulting in activation of components of intracellular Smad-dependent and Smad-independent pathways. In the Smad-dependent pathway, receptor-regulated Smads (R-Smads) become phosphorylated at their C termini via type I receptor kinase activity, complex with Smad4, and accumulate in the nucleus where they act as transcriptional co-regulators (2,3). Initially it was thought that TGF-␤s signal exclusively through TGF-␤ type I and II receptors (TGF␤RI and TGF␤RII) to activate TGF-␤ R-Smads 2 and 3 and similarly BMPs signal through specific Type I and II receptors to activate BMP R-Smads 1, 5, and 8 (4). However, a recent report showed that at least in certain specific cell lines, TGF-␤s can also signal via mixed receptor complexes, resulting in activation of BMP R-Smads (5).
In addition to C-terminal phosphorylation at a conserved SSXS sequence, more recent research has shown that R-Smads can also be phosphorylated in the linker region between their Mad homology domains, which is important for both full activation and cessation of activation (6,7). Specific residues become phosphorylated as the result of distinct pathways: antagonists, such as EGF, which enable the activated R-Smads to remain in the cytoplasm and be degraded (8); and agonistinduced linker phosphorylation that is required for maximal transcriptional activity of R-Smad-Smad4 complexes, as well as for rapid turnover of corresponding R-Smads (7,9). In this "action turnover switch" model, the Thr/Ser residues adjacent to proline-rich sequences in R-Smads (Thr-220 in Smad2) are phosphorylated by nuclear Cdks creating a docking site for transcriptional co-regulators (Pin1 in the case of TGF-␤ R-Smads and Yap in the case of Bmp R-Smads). Moreover, it has been suggested that this phosphorylation may enable subsequent phosphorylation of other linker region residues, such as Ser-250 in Smad2, providing a docking site for ubiquitin ligases, such as Nedd4L, targeting R-Smads for proteasomemediated degradation (9,10).
Although the canonical pathway tends to be treated as the more important, recent studies suggest that the Smad-independent TGF-␤ signaling plays a more significant role in vivo than previously thought (16,17). In a mouse model of Marfan's syndrome in which Tgf-␤ signaling is elevated, the aortic aneurysm phenotypes are made worse by reducing the canonical pathway component Smad4, and the phenotypes are made less severe by attenuation of noncanonical pathways (16). These results show not only that the noncanonical pathway is of importance in disease mechanisms but also that the two pathways somehow interact with one another. Mutations in TGF-␤ type I and type II receptors (TGFBR1 and TGFBR2) associated with Marfan's syndrome and Loeys-Dietz syndrome lead to an increase in TGF-␤ signaling activity in the absence of normal canonical signaling (18). The importance of maintaining the correct balance between the pathways for normal developmental events is implied by the presence of craniofacial defects including cleft palate in patients suffering from Loeys-Dietz syndrome (19) and in a mouse model when Tgfbr2 is genetically removed from the neural crest cell population (which forms the principal source of mesenchymal cells in the secondary palate precursor structures, the palatal shelves), resulting in cleft palate in the presence of excessive TGF-␤-induced noncanonical signaling (17).
In light of the important role played by the Smad-independent pathway implied by these studies, as well as the well established importance on Tgf-␤ and Bmp signaling in normal craniofacial development in mouse models (17), we examined the role of Tak1 in craniofacial neural crest development by deleting Tak1 function in premigratory neural crest cells using the Wnt1-Cre driver line. The mutant mice display hypoplastic facial structures and cleft palate, which is caused by a delayed palatal shelf elevation. We found that Tak1 is required for appropriate activation of both p38 Mapk (noncanonical pathway) and TGF-␤/Bmp R-Smads (canonical pathway) in the neural crest-derived craniofacial ecto-mesenchyme. We also show that Tak1 deficiency results in attenuated TGF-␤ R-Smad linker region phosphorylation and that Tak1 kinase mediates both distinct and overlapping agonist-induced transcriptional responses. Collectively, these results imply that in neural crestderived mesenchymal cells, Tak1 mediates both canonical and noncanonical arms of the TGF-␤ superfamily signaling.

EXPERIMENTAL PROCEDURES
Mice-Tak1-flox mice were generated by flanking a critical exon 2 with asymmetric loxP sites (Lox66 and Lox71) in opposing orientations (see Fig. 1; details will be described elsewhere). Wnt1-Cre (from the Jackson Laboratories) and Tgfbr1 FX (kindly provided by S. Karlsson) mice have been described earlier (20,21). Tak1 FXϩ /Wnt1-Cre ϩ male mice were crossed with Tak1 FXFX female mice to obtain timed pregnancies. The presence of a vaginal plug was designated as embryonic day 0 (E0). DNA for genotyping was prepared from yolk sac or from tail tissue using DirectPCR lysis reagents (Viagen Biotech). Mouse lines were maintained in mixed genetic backgrounds. All experiments involving the use of animals were approved by the Institutional Animal Use and Care Committee at the University of Michigan at Ann Arbor.
Real Time Quantitative PCR-Total RNAs were isolated, and cDNAs were synthesized as outlined above. Real time quantitative PCR experiments were carried out using Universal Probe Library-based assays (Roche Applied Science) with gene-specific primer sequences generated by the manufacturer's online algorithm and TaqMan Universal PCR master mix (Applied Biosystems). 30-l assays were quantified using the ABI7300 PCR and detection system (Applied Biosystems) and analyzed using 7500 System v1.2.2 software.
Histology, in Situ Hybridization, Cell Death, and Proliferation Assays-For histological analyses, the tissues were processed, sectioned, and stained by hematoxylin and eosin according to standard protocols. For whole mount in situ hybridization, the Tak1 Signaling in the Craniofacial Neural Crest tissues were fixed in 4% buffered formaldehyde for 12-16 h and dehydrated though a graded methanol series (20,50,70,95, and 100%) containing PBST. Antisense RNA probes were synthesized with NTP digoxigenin RNA labeling mix (Roche Applied Science) following the manufacturer's instructions. Probe templates for Shox2, Tbx22, and Shh were obtained from Y.-P. Chen (22), R. Jiang (23), and S. Bellusci (24), respectively. A probe template for Tgfb3 was prepared as described (25). Apoptotic cells were detected using a TUNEL assay (Dead End from Promega) following the manufacturer's instructions. For cell proliferation analyses, cell proliferation labeling reagent (RPN201; Amersham Biosciences) was used. BrdU-positive cells were detected using anti-BrdU antibody (RPN202; Amersham Biosciences). Fluorescent images were viewed on an Olympus BX51 microscope and documented using an Olympus DP71 camera.
MicroCT Analyses-Specimens were embedded in 1% agarose, placed in a 19-mm-diameter tube, and scanned over the entire length of the skull using a microCT system (CT100; Scanco Medical, Bassersdorf, Switzerland). The scan settings were: voxel size 10 m, medium resolution, 55 kVp, 109 A, 0.5 mm AL filter, and integration time 500 ms. Images were created using the manufacturer's evaluation software and a fixed global threshold to segment bone from non-bone. Skull shape was analyzed by measuring the distance from the supraoccipital bone to the anterior end of the frontal bone (length) and from the top of the parietal bone to the cranial base (height) and calculating the height/length ratio. Mandibular length was measured as shown in Fig. 1.
Primary Craniofacial Mesenchymal Cell Cultures-Mandibular and maxillary processes of the first mandibular arch and prefusion palatal shelves were dissected from E11.0 and E14.0 embryos, respectively. The cultures were established as described by Thomas et al. (27). The neural crest origin and purity of cultures was confirmed by establishing cultures of embryos that carried both the Wnt1-Cre transgene and were positive for the R26-lacZ reporter (Fig. 3B). ␤-Galactosidase staining was carried out as described (28). After 7-10 days, the cells were stimulated with growth factors (10 ng/ml TGF-␤2 or 100 ng/ml BMP2) for 0, 10, or 40 min and harvested for Western blot analyses. Chemical inhibitors flavopiridol (Cdk inhibitor; 1 or 10 M; Sigma-Aldrich), UO126 (Erk1/2 (and Jnk) inhibitor; 10 M; Sigma-Aldrich), SB431542 (TGF␤RI inhibitor; 10 M; Sigma-Aldrich), 5Z-7-oxozeaenol (Tak1 inhibitor; 2 M; eMolecules), and SB202190 (p38 Mapk inhibitor; 10 M; Sigma-Aldrich) were added to cells 60 min prior to TGF-␤2 or BMP2 addition. The cells were not starved of serum prior to these treatments. Some cultures were transduced with replication-deficient recombinant adenoviral preparations using a multiplicity of infection of 1000. Ad-Gfp and Ad-Cre (titer, 4 ϫ 10 12 viral particles/ml) were obtained from the University of Michigan Biomedical Research Core Facility. For immunofluorescence, cultured cells were fixed in 4% buffered formaldehyde (5 min) and stained using anti-Smad2 antibody (antibody 5339; Cell Signaling). Binding was detected using Alexa Fluor 594 goat anti-rabbit secondary antibody (Invitrogen), and the slides were mounted in Vectashield with DAPI (Vector Labs Inc.). Fluorescent images were viewed and documented as outlined above. In some experiments, nuclear and cytoplasmic fractions were separated using the NE-PER nuclear and cytoplasmic extraction kit (kit 78833; Thermo Scientific) according to the manufacturer's instructions.
Neural Crest Stem Cell Cultures-The neural crest stem cell line O9-1 was obtained from M. Ishii. Undifferentiated cells were cultured in the presence of leukemia inhibitory factor and basic FGF in SNL cell-conditioned medium as described (29).
Roller Bottle Organ Cultures-Heads from embryos were collected at E13.5, and mandibles, tongues, and brains were removed in PBS. The resulting mid-face samples were cultured for 24 -48 h at 37°C in roller bottles (60 rotations/min) in serum-free BGJb medium without penicillin and streptomycin. The bottles were gassed at the beginning of the culture and every 12 h by gently bubbling the medium for 2 min with O 2 /CO 2 (95%/5%). The palatal cultures were fixed, sectioned, and stained as described above.
Microarray Analyses-Palatal mesenchymal cells were isolated and aliquoted to 4 groups as shown in Table 1. Confluent cultures were incubated in the presence of inhibitors (TGF␤RI inhibitor SB431542, 10 M; Tak1 inhibitor 5Z-7-oxozeaenol, 2 M) for 1 h and then stimulated with TGF-␤2 (10 ng/ml) for 2 h. Total RNAs were then isolated using an RNeasy kit (Qiagen). Concentration and purity were determined by spectrophotometry, and RNA integrity was confirmed by running aliquots on 1% agarose gel. 500 ng of total RNA of each sample was amplified and biotin-labeled according to the GeneChip 3Ј IVT expression kit user manual (Affymetrix). Microarray analysis was performed with the GeneChip mouse genome 430 2.0 array (Affymetrix). The biotin-labeled RNAs were fragmented, and 10 g of each was hybridized to a GeneChip at 45°C for 16 h. Labeled bacterial RNAs of known concentrations were spiked in hybridization to generate an internal standard and to allow normalization between the chips. The chips were washed and stained with streptavidin R-phycoerythrin (Molecular Probes). After scanning the chips by GeneChip Scanner 3000 7G System, the data were analyzed by Affymetrix GeneChip-related software packages GCOS, Data Mining Tool, and Affymetrix web database "NetAffx." Microarray data were submitted to the GEO repository under accession number GSE45491.

Neural Crest-specific Tak1 Mutants Display Mandibular Hypoplasia and Cleft
Palate-A recent study demonstrated that haploinsufficiency of Tak1 in neural crest cells rescued the cleft palate phenotype of Tgfbr2 mutant mice (17). However, the role of Tak1 itself in craniofacial development is currently poorly known. To address this question, we crossed mice homozygous for the floxed Tak1 allele (Tak1 FXFX ; for details, see Fig. 1A) with transgenic Wnt1-Cre mice that also were heterozygous for the floxed Tak1 allele (Tak1 FXϩ /Wnt1-Cre ϩ ). The expected Mendelian proportion of Tak1 FXFX /Wnt1-Cre ϩ mice (hereafter referred to as Tak1/Wnt1-Cre mutants) survived to birth (n ϭ 16 of 66) but then died within 24 h.
We used micro-CT analysis to compare the structure of control and Tak1/Wnt1-Cre mutant littermate heads at P0 in detail. The craniofacial skeleton appeared generally stunted in mutants (Fig. 1C). Many of craniofacial bones are largely derived from neural crest cells, including cranial base, mandible, maxilla, and frontal skull (29). Although the average lengths Left panel, exon 1-specific sense and exon 3-specific antisense primers result in amplification of 261-bp wild-type and 150-bp mutated PCR products, respectively. Exon 1-specific sense and exon 4-specific antisense primers result in amplification of 448-bp wild-type and 337-bp mutated PCR products, respectively. C, lateral and superior CT images of control (Tak1 FX/ϩ /Wnt1-Cre Ϫ ) and mutant (Tak1 FX/FX /Wnt1-Cre ϩ ) heads at P0. The yellow arrows in the lateral images depict the length of the mandible. The yellow arrows in the superior views depict the length of the maxilla. The red arrows in the superior images depict the length of the cranial base, and the green arrows illustrate palatine bones in a control (fused palate) and mutant (cleft palate). The bar graph shows lengths of the cranial base, maxilla, and mandible of controls (blue columns) and mutants (red columns) (n ϭ 3). D-F, histological comparison of palatal phenotypes of controls and Tak1/Wnt1-Cre mutants: frontal sections at anterior, middle, and posterior levels at E14.5 (E), E18.0 (F), and the level of the first molar at E13.5 (D). The black arrows point to tips of palatal shelves, the black arrowheads point to the midline seam (in a control), and the black vertical lines depict the distance between the tongue and nasal septum. G, left-hand panel shows sections from two independent control and mutant samples at the level of the first mandibular molar. The green highlighting illustrates the area measured to compare palatal shelf size between controls and mutants. Middle panel, bar graph shows relative quantification of palatal shelf (PS) size in controls (C, blue columns) and mutants (M, red columns) (n ϭ 3). Right panel, bar graph summarizes results from BrdU incorporation assays on controls (blue columns) and Tak1/Wnt1-Cre mutants (red columns) palatal shelves at E13.5 (n ϭ 4). Error bars, S.E. *, p Ͻ 0.05. NS, nasal septum; P, palatal shelf; SP, secondary palate.

Tak1 Signaling in the Craniofacial Neural Crest
of control and Tak1/Wnt1-Cre mutant cranial bases did not differ, those of the mandible and maxilla were ϳ20 and 5% shorter, respectively, in the mutants (Fig. 1C). The skulls in mutants were also shorter than in controls (length versus height ratio 1.74 Ϯ 0.04 (average Ϯ S.E.) and 1.49 Ϯ 0.02 in controls and mutants, respectively; n ϭ 3). The mutants also displayed cleft secondary palate with high penetrance (13 of 16, 80%).
To investigate the mechanism underlying the palatal defect, we collected mutant and control embryos at defined time points and analyzed palatal phenotype in detail by histology ( Fig. 1, D-F). At E13.5, the palatal shelves in both control and mutants were still growing vertically toward the floor of the mouth, parallel to the sides of, and separated by, the tongue, which was almost square in cross-section (Fig. 1D). By E14.5 (Fig. 1E), in controls the tongue had changed shape in crosssection, becoming wider than it was tall. It was no longer interposed between the palatal shelves, which had elevated to point toward one another parallel to the top of the tongue, and commenced fusion with one another along the midline of the oral cavity in the mid-and posterior regions, although a "seam" of adhered epithelia from each shelf still remained. In contrast, in Tak1/Wnt1-Cre mutants, the tongue remained interposed between the palatal shelves, lying close to the nasal septum, and was still relatively square in cross-section. The shelves had failed to elevate very much. By E18 (Fig. 1F), in controls the secondary palates had fused completely along their entire length: the seam formed where their epithelia had met in the midline had disappeared, so their mesenchymal cores were confluent, and differentiation into cartilage commenced. In most Tak1/Wnt1-Cre mutants examined, secondary palate formation had failed, leaving a cleft from anterior to posterior. Although the tongue was no longer completely interposed between the palatal shelves, which had elevated, the shelves had failed to adhere to or fuse with one another, although some cartilage formation had occurred. Although now above the tongue in anterior and mid-positions, the underside of the shelves and the top of the tongue were very close compared with the same structures in controls.
Because components of the head were smaller in mutants than controls at birth, we examined the relative size of palatal shelves in littermates at E13.5 and found that shelves in mutants were ϳ20% smaller in cross-sectional area than those in controls (Fig. 1G, left panel). BrdU incorporation assays on prefusion palatal shelf sections from littermates showed a lower average value in mutant versus control cells at the anterior level (not significantly different), but no difference at the mid and posterior levels (Fig. 1G, right panel). To determine whether the smaller mutant shelf area at E13.5 was the result of an earlier lower cell proliferation rate, we performed the same assay at E11 on mandibular primordia (first pharyngeal arch mesenchyme), but no significant difference between control and mutant tissues was detected (data not shown). No differences in cell death were detected at E11 or E13.5 (data not shown).
Because Tak1/Wnt1-Cre mutants displayed a distinct reduction in craniofacial size at birth (Fig. 1), likely resulting in a hypoplastic oral cavity, and the position and shape of the tongue appeared abnormal in mutants during palatogenesis, we wondered whether the failure in palatal shelf elevation could be caused by the failure of the tongue to "descend" appropriately. To address this, at E13.5 (when palatal shelves are still growing vertically) we cultured mouse embryonic heads from which we had removed the mandible and tongue. This was performed in roller bottles under chemically defined conditions. In both control and mutant samples, sufficient elevation of palatal shelves occurred after 24 h that, after a further 24-h culture, fusion occurred, with equivalent mesenchymal confluence regardless of genotype (Fig. 2, A and B). This suggests that in Tak1/Wnt1-Cre mutants the tongue obstructed palatal shelf elevation in vivo, contributing to the pathogenesis of cleft palate in Tak1/ Wnt1-Cre mutants in vivo (44).
Previous studies have shown that TGF-␤ superfamily signaling is involved in the anterior-posterior patterning of the secondary palate (22, 30 -32). Therefore, we compared expression patterns of established anterior (Shox2) and posterior (Tbx22) mesenchymal markers between Tak1/Wnt1-Cre mutant and control littermates using whole mount in situ hybridization. These experiments did not demonstrate detectable differences in Shox2 and Tbx22 expression patterns or intensities between controls and mutants (Fig. 2C), suggesting that in Tak1/Wnt1-Cre mutants, the anterior-posterior patterning is not grossly affected.
Mesenchymal Bmp signaling has been shown to be important for maintenance of Shh expression in the rugae of the palatal epithelium (30), which in turn plays a critical role in regulation of growth and patterning along the palatal oro-nasal axis (33,34). Our experiments demonstrate that, unlike palatal mesenchyme-specific Bmpr1a mutant shelves (35), there was no reduction in Shh expression in the primary palate of Tak1/ Wnt1-Cre mutants. However, whereas the amount of Shh expression in individual ruga was not decreased, at the time of palatal fusion (E14.5-E15.5), a marked delay in rugal formation suggests that the rate of expansion of the secondary palate was reduced (Fig. 2D).
In summary, Tak1 deficiency in the craniofacial neural crest resulted in facial (particularly mandibular) and palatal shelf hypoplasia and cleft palate. Palatal shelf elevation was delayed, and the tongue was abnormal in position and shape in mutants immediately prior to the normal stage of palatal fusion. No difference between control and Tak1/Wnt1-Cre mutants in anterior-posterior patterning of prefusion palatal shelves was detected by marker gene analysis, but a delay in ruga pattern formation suggests abnormal morphogenesis or maturation there. In rolling culture, removal of the tongue and mandible resulted in mutant palatal shelf elevation, and fusion occurred in both control and mutant heads.
Both Canonical and Noncanonical TGF-␤ Superfamily Signaling Pathways Are Affected in Tak1/Wnt1-Cre Mutants-Neural crest cells make a contribution to all the craniofacial structures identified as showing abnormalities in Tak1/Wnt1-Cre mutants, including palatal shelves, tongue, and mandible. Of these, the palatal shelf mesenchyme is the most uniformly composed of neural crest cells. To identify signaling processes influenced by Tak1 deficiency in vivo, we harvested prefusion palatal shelves from control and Tak1/Wnt1-Cre embryos and analyzed their protein for Tak1, phosphorylation of Tak1, and its downstream signaling target p38 Mapk, the TGF-␤ signaling Tak1 Signaling in the Craniofacial Neural Crest MAY  target Smad2, and BMP signaling targets Smad1/5/8 (Fig. 3A). The level of Tak1 and Tak1 C-terminal phosphorylation at Ser-412 was significantly less in Tak1/Wnt1-Cre mutant samples, demonstrating an efficient Cre-induced recombination in the Tak1 locus. The level of p38 Mapk phosphorylation was also markedly and consistently less in these mutant palatal shelves when compared with corresponding control samples. Similarly, C-terminal Smad2 phosphorylation was consistently lower in mutant shelves, whereas there were no differences in Smad2 protein levels. C-terminal Smad1/5/8 phosphorylation was lower on average in mutant samples but not significantly so. These results show that Tak1 deficiency leads to decreased signaling activity in both noncanonical and canonical TGF-␤ signaling pathways.
Tak1 Mediates C-terminal Smad Phosphorylation in Craniofacial Mesenchymal Cells-To gain a deeper insight into processes regulated by Tak1-mediated signaling in postmigratory neural crest cells, we established control and Tak1/Wnt1-Cre mutant craniofacial mesenchymal primary cultures from both prefusion palatal shelves (at E14) and first pharyngeal arches, which give rise to maxilla and mandible (at E11), and analyzed their ability to respond to TGF-␤ stimulation (Fig. 3, B and C). As expected, control mesenchymal cells isolated from the secondary palate showed an increase in Tak1 Ser-412 phosphorylation in response to TGF-␤ stimulation and beyond a clearly detectable base-line Tak1 phosphorylation level (Fig. 3C). In Tak1/Wnt1-Cre cells, there was no detectable Tak1 phosphor-ylation, with or without TGF-␤ stimulation. A modest increase in phosphorylation of Smad-independent pathway targets p38 Mapk, Jnk, and Erk1/2 was induced by TGF-␤2 in control cells, but in mutant cells no such change was observed (Fig. 3C). TGF-␤2 stimulation induced C-terminal phosphorylation of both TGF-␤ Smads (Smad2) and BMP Smads (Smad1/5/8) in control palatal mesenchymal cells. In mutant cells, Smad2 C-terminal also became phosphorylated in response to TGF-␤2 stimulation, but less so than in control cells (Fig. 3, B and C), and BMP Smads did not show detectable C-terminal phosphorylation. In all samples, Smad2 protein levels remained stable for 90 min after TGF-␤ stimulation (Fig. 3B). Similar results were obtained using cells isolated from first pharyngeal arches (which also contribute to mandible and tongue) at E11, using a neural crest stem cell line in conjunction with the Tak1 kinase inhibitor (5Z-7-oxozeaenol) (Fig. 3F and data not shown).
In summary, the relative pattern of phosphorylation of downstream targets shown by control and mutant palatal shelf primary culture cells in response to TGF-␤ stimulation is similar to that found between control and mutant E14 palatal shelves recovered in vivo. Because Tak1 has also been shown to be involved in BMP signaling, we stimulated both the control and Tak1/Wnt1-Cre mutant ecto-mesenchymal cells with BMP2 and monitored the phosphorylation of signaling molecules as above (Fig. 3D). Unlike TGF-␤2 and as expected, BMP2 was not able to induce Tak1 phosphorylation at Ser-412 in control cells, nor did mutant cells show detectable phospho-Tak1

Tak1 Signaling in the Craniofacial Neural Crest
levels. Control cells were responsive to BMP2 stimulation in their progressive increase in C-terminal BMP-Smad (Smad 1/5/8) phosphorylation, but Smad2 phosphorylation was not altered. In mutant cells, base-line C-terminal Smad1/5/8 phosphorylation levels were lower, and their responsiveness to BMP2 stimulation was reduced, when compared with control cells. Collectively, the results of these experiments suggest that in neural crest-derived mesenchymal cells, Tak1 contributes more to TGF-␤ than BMP-induced C-terminal phosphoryla-tion of the corresponding R-Smads. BMP2 induced progressive p38 Mapk and Jnk phosphorylation in controls cells, but no increase was detected in mutant cells.
To provide additional evidence for the role of Tak1 in C-terminal phosphorylation of Smad2, we transduced palatal mesenchymal cells carrying the homozygous Tak1-flox allele with recombinant adenoviruses expressing the Cre recombinase and analyzed the transduced and control cells (transduced with AdGfp) for Tak1 and Smad2 phosphorylation (Fig. 3E). These Co, control sample; Mu, Tak1/Wnt1-Cre mutant sample. The bar graph shows relative quantification of pTak1, p38 Mapk, pSmad2-C, and pSmad1/5/8-C in controls (blue columns) and mutants (red columns) (normalized to ␤-actin; n ϭ 4). B, primary palatal mesenchymal cell cultures were established (see "Experimental Procedures") from prefusion palatal shelves (at E14.0) of mouse embryos carrying both the Wnt1-Cre transgene and the R26R lacz reporter (top left). Cells stained blue are derived from the neural crest. The black arrows point to a few negatively staining cells. Scale bar, 200 M. A representative immunoblot of primary palatal mesenchymal cells stimulated with TGF-␤2 and analyzed for C-terminal Smad2 phosphorylation is shown. Top right, the diagram illustrates the differences in amount of normalized Smad2-C phosphorylation between control (blue data points and blue line) and Tak1/Wnt1-Cre mutant (red data points and red line) cells. The values were normalized to Smad2. C, representative immunoblot from primary palatal mesenchymal cells stimulated with TGF-␤2 (10 ng/ml), and protein lysates were analyzed for Tak1, C-terminal Smad1/5/8, C-terminal Smad2, total Smad2, p38 Mapk, Jnk, and Erk1/2 phosphorylation. The bar graphs show relative quantification of pTak1, p38 Mapk, pSmad2-C, pJnk, and Smad2 in controls (blue columns) and mutants (red columns) (normalized to ␤-actin; n ϭ 3). D, representative immunoblot of primary palatal mesenchymal cells stimulated with BMP2 (100 ng/ml) and analyzed for Tak1 phosphorylation, C-terminal Smad1/5/8 phosphorylation, total Smad1, C-terminal Smad2 phosphorylation, total Smad2, p38 Mapk phosphorylation, and Jnk phosphorylation. The bar graphs show relative quantification of pTak1, p38 Mapk, pSmad1/5/8-C, and total Smad1 in controls (blue columns) and mutants (red columns) (normalized to ␤-actin; n ϭ 3). Error bars, S.E. *, p Ͻ 0.05. E, representative immunoblot of primary Tak1 FX/FX palatal mesenchymal cells transduced with AdGfp (control) or AdCre (mutant). The transduced cells were stimulated with TGF-␤ and analyzed for Tak1 phosphorylation, C-terminal Smad2 phosphorylation, total Smad2, and ␤-actin. Moi, multiplicity of infection. F, representative immunoblot of O9-1 neural crest stem cells stimulated with TGF-␤2 in the presence or absence of the Tak1 kinase inhibitor 5Z-7-oxozeaenol (5Z) and analyzed for C-terminal Smad2 phosphorylation, total Smad2, and ␤-actin. MAY 10, 2013 • VOLUME 288 • NUMBER 19 results confirm that the reduction in C-terminal phosphorylation of Smad2 is an immediate result of Tak1 deficiency rather than just a long term adaptation to severely reduced Tak1 levels in our Tak1/Wnt1-Cre mutants.

Tak1 Signaling in the Craniofacial Neural Crest
To conclude, we show that Tak1 is required for appropriate activation of both canonical and noncanonical TGF-␤ and BMP signaling pathways in neural crest-derived mesenchymal cells. Moreover, in the presence of Tak1, TGF-␤2 was able to induce activation of both TGF-␤ and BMP R-Smads, i.e. Smad2 and Smad1/5/8, respectively.
Agonist-induced Linker Region Phosphorylation Is Affected in Palatal Mesenchymal Cells Deficient in Tak1 Mutants-Several studies have shown that in R-Smads the intervening sequence that links the DNA-binding domain and transcriptional domain, the so-called linker region, becomes phosphorylated by a set of divergent stimuli (including FGFs, EGFs, and stress signals) via Mapks and also by members of the TGF-␤ superfamily (36 -38). This agonist-induced linker region phosphorylation (ALP) at Thr-220 in Smad2 (Thr-179 in Smad3) located immediately upstream of a PY motif is of particular interest, because it was recently suggested that, in addition to priming Smads for turnover, phosphorylation of this threonine residue is required for a full transcriptional activity of Smad complexes (7,9). Therefore, we stimulated both control and Tak1-deficient primary mesenchymal cells (from both the first pharyngeal arch and the palate) with TGF-␤2 and analyzed phosphorylation of these key residues (Smad2-L(T220) and Smad3-L(T179)) and of Smad2-L(S250) using phospho-residuespecific antibodies. Our results demonstrate that in control cells TGF-␤2 induced efficient phosphorylation of the Smad2 linker region at Ser-250 and Thr-220 and at Thr-179 in Smad3, whereas cells deficient in Tak1 failed to show comparable induction of ALP (Fig. 4A). As before, practically identical results were obtained using cells isolated from first pharyngeal arches at E11.0 (data not shown). Moreover, O9-1 neural crest stem cells showed similar increases in TGF-␤2-induced linker region phosphorylation at Smad2-L(T220) and Smad3-L(T179) and showed that this induction was largely inhibited by 5Z-7oxozeaenol (Fig. 4A, lower panel).
To investigate whether these in vitro results were reflective of palatal shelf mesenchyme in vivo, we prepared protein lysates directly from prefusion palatal shelves of control and Tak1/Wnt1-Cre mutant embryos and analyzed them for the Smad2/3 linker region phosphorylation. Endogenous levels of Smad2 linker phosphorylation at Thr-220 and Ser-250 and Smad3 phosphorylation at Thr-179 were all significantly lower in mutant palatal samples than in those of controls (Fig. 4B).
It has previously been shown that in several established cell lines, e.g., HEK293T and HaCaT, Thr-220 in the linker region of Smad2 is directly phosphorylated by nuclear cyclin-dependent kinases Cdk8 and Cdk9 (7,9). Therefore, we examined the effect of Tak1 inactivation on subcellular localization of total Smad2 both on palatal mesenchymal cells (harvested at E14) (Fig. 4C) and on first pharyngeal arch mesenchymal cells (harvested at E11; data not shown). In both cell types, TGF-␤2induced Smad2 nuclear accumulation was effectively inhibited by the Tak1 kinase inhibitor (5Z-7-oxozeaenol) and by the TGF␤RI kinase inhibitor (SB431542) (Fig. 4C and data not  shown). Tak1 kinase inhibitor also inhibited both TGF-␤2-induced Smad2-C-terminal and linker region phosphorylation at Smad2-L(T220), Smad3-L(T179) and Smad2-L(S250) (Fig. 5A). Certain Tak1-mediated ALP was also dependent on the presence of the TGF␤RI protein and the kinase activity of TGF␤RI, because both the Cre-mediated Tgfbr1 deletion and TGF␤RI inhibition by the SB431542 could attenuate TGF-␤2-induced phosphorylation at Smad2-L(T220)/Smad3-L(T179) (Fig. 5, A  and B). As expected, agonist-induced phosphorylation of Smad2 at the C terminus was also inhibited by the physical loss of Tgfbr1 and by SB431542 (Fig. 5, A and B). In palatal mesenchymal cells, flavopiridol (Cdk inhibitor) inhibited ALP at Thr-220 and Ser-250 (Smad2) (Fig. 5A), whereas UO126 (Erk1/2 and Jnk inhibitor) could effectively prevent TGF-␤2-induced Smad2-L(S250) phosphorylation but could only weakly inhibit ALP at Thr-220 (Fig. 5A). The effect of SB202190 (p38 Mapk inhibitor) on phosphorylation at Smad2-L(T220) was very similar to that of UO126 (data not shown). Our cell fractionation experiments confirmed that most of the Smad2 molecules with TGF-␤2-induced phosphorylation at Thr-220 could be detected in the nucleus and that this phosphorylation was effectively inhibited by flavopiridol (Fig. 5C). In summary, our results show that in craniofacial mesenchymal cells Tak1 deficiency led to reduced agonist-induced C-terminal phosphorylation and decreased nuclear accumulation of R-Smads and that agonist-induced linker phosphorylation of Smad2 both at Thr-220 (which is at least partially mediated by Cdks) and at Ser-250 (mediated by Erk1/2 and/or Jnk) was dependent on the functional Tak1 protein (Fig. 5, D and E).
Tak1 as a TGF-␤ Signal Transducer in Neural Crest-derived Facial Mesenchymal Cells-Previous studies have demonstrated that TGF-␤-induced Tak1 activation is dependent on the presence of TGF␤RI protein and is mediated by an adaptor protein Traf6 (11,12). However, whether or not this activation is dependent on the TGF␤RI kinase activity is still controversial (11,12). As shown above, the presence of either TGF␤RI or its kinase activity is sufficient for agonist-induced Tak1-mediated linker region phosphorylation of Smad2 in ecto-mesenchymal cells (Fig. 5, A and B). To better define the relationship between Tak1 kinase-mediated and TGF␤RI kinase-mediated transcriptional responses, we compared expression profiles of TGF-␤2responsive genes in primary palatal mesenchymal cells in the presence of inhibitors for either TGF␤RI kinase or Tak1 kinase using a genome-wide transcriptomic analysis (Fig. 6). Of TGF-␤2-responsive genes (Ͼ2-fold change; p Ͻ 0.05), 58 were dependent on TGF␤RI kinase, 27 were dependent on Tak1 kinase and 18 were dependent on both (Ͼ2.5-fold change; p Ͻ 0.05) (Fig. 6A). Among the up-regulated genes (120 genes; Ͼ2-fold change; p Ͻ 0.05), 47 were TGF␤RI kinase-dependent, 20 were Tak1 kinase-dependent, and 12 were dependent on both (Ͼ2.5-fold change; p Ͻ 0.05) (Fig. 6B). Among the downregulated genes (45 genes; Ͼ2-fold change; p Ͻ 0.05), 11 were TGF␤RI kinase-dependent, 7 were Tak1 kinase-dependent, and 6 were dependent on both (Ͼ2.5-fold change; p Ͻ 0.05) (Fig. 6C). Differential expression of several Tak1 kinase/ TGF␤RI kinase-dependent genes was confirmed by real time RT-PCR (Fig. 5D). When more stringent criteria were used and

Tak1 Signaling in the Craniofacial Neural Crest
only the genes that displayed Ͼ3-fold expression change when stimulated with TGF-␤2 were analyzed, 10 genes were dependent on both TGF␤RI and Tak1 kinase activities, whereas 8 genes and 1 gene were dependent on only TGF␤RI or only Tak1 kinase activities, respectively (data not shown).

Tak1 Signaling in the Craniofacial Neural Crest
with the findings made in mesenchymal cell culture assays in vitro.
To conclude, TGF␤RI and Tak1 kinases mediated both distinct and overlapping transcriptional responses in palatal mesenchymal cells, when stimulated by TGF-␤2. In this cell type, agonist-induced Tak1 activation could occur independently of TGF␤RI kinase activity.

DISCUSSION
Although Tak1, a member of the Mapk kinase kinase subfamily, was originally shown to mediate TGF-␤ signaling pathways (39), its physiological role in TGF-␤ superfamily signaling in vivo, particularly during embryogenesis, is still poorly elucidated. Here we show that Tak1 plays a critical nonredundant role in craniofacial development, and that, in embryonic neural

Tak1 Signaling in the Craniofacial Neural Crest
crest-derived craniofacial cells, Tak1 is required not only for Smad-independent TGF-␤ superfamily signaling but also for maximal ligand-induced C-terminal and linker phosphorylation of R-Smads.
Role of Tak1 in Craniofacial Development-Previous studies have shown that TGF-␤ and Bmp signaling plays an essential role in the neural crest-derived palatal mesenchyme (35,40,41). Of particular note is a recent report demonstrating that cleft palate in Tgfbr2/Wnt1-Cre mutants results from TGF-␤2induced activation of an alternative pathway via TGF␤RI and TGF␤RIII that induces activation of Tak1 and hence p38 Mapk (17). This aberrant activation results in attenuated expression of Fgf9 and Pitx2, leading to reduced mesenchymal cell proliferation and cleft palate (42). These results raised the questions of what role neural crest-expressed Tak1 normally plays in craniofacial development and in the activation of signaling pathway components downstream of TGF-␤2 signaling.
Here we report that Tak1 is required for normal facial and mandibular growth and palatogenesis in vivo and that it mediates both Smad-dependent and Smad-independent TGF-␤ and Bmp signaling in neural crest cells in vitro. The overall craniofacial phenotypes of Tak1/Wnt1-Cre mutants were relatively mild when compared with those of the corresponding Tgfbr1, Tgfbr2, or Bmpr1a mutants. Unlike in Tgfb receptor mutants, palatal mesenchymal cell proliferation in general was only slightly affected in Tak1 mutants, because we did not detect any significant differences at the time points examined. Although overactivation of Tak1 in the Tgfbr2/Wnt1-Cre mutants led to a reduction in cell proliferation (16), removal of Tak1 in our Tak1/Wnt1-Cre mutants resulted in moderate hypoplasia. Because we could not detect increased apoptosis between embryonal days 11-14, generalized hyperplasia in NC-derived tissues as the result of the absence of Tak1 does not seem to occur.
Normal formation of the secondary palate requires distinct developmental events; not just formation and growth of two palatal shelves but their elevation, physical contact with one another, fusion, loss of epithelial seam, and differentiation into cartilage and bone. Of these steps, shelf elevation and patterning (ruga formation) showed developmental delay in Tak1Wnt1Cre mutants. It was recently shown that a Turingtype reaction-diffusion mechanism establishes the normal pattern of rugae (43). Although the delay in ruga formation may indicate an intrinsic defect in palatogenesis, it also is possible that a failure of the tongue to descend disturbs the shelf tissue proportions, altering the distribution of signals for ruga formation and slowing their increase in number. An essential role for Tak1 in palatal shelves themselves in the processes of palatal shelf elevation onwards is hard to define because these processes occurred indistinguishably in control and Tak1/Wnt1-Cre mutant heads in rolling culture in vitro when tongue and mandible were absent, but their morphology looked different from normal palatal shelves in vivo (Fig. 1B). This result, along with histological observations, suggests that the failure of the tongue to descend sufficiently in mutants most likely contributes to the cleft palate phenotype in vivo by physically impeding shelf elevation. Consistent with these findings, Song et al. have recently shown that Tak1 plays an important role in tongue development by controlling Fgf10 expression (44).
At the protein level, Tak1/Wnt1-Cre mutant tissues showed reduced levels of activation of both Smad-independent and -dependent pathways rather than an up-regulation of Smad-dependent pathway signaling. As our studies on primary culture cells also showed, Tak1 functions as a modulator of signal strength rather than an "on/off switch" for the Smad-independent pathway in the NC-derived craniofacial mesenchyme. Our results do not exclude that altered TGF-␤/Bmp-independent Tak1-mediated signaling in these cells may also contribute to the relatively mild phenotype.
Role of Tak1 in TGF-␤-induced Signaling Events-Our studies of the dependence of TGF-␤2-induced signaling on Tak1 in embryonic neural crest cells have revealed important details not only of phosphorylation of components of signaling pathways but of their cell type-dependent variation. We show that in NC-derived craniofacial mesenchymal cells, Tak1 mediates both TGF-␤ and BMP-induced activation of their corresponding R-Smads. Shim et al. (45) showed that in chondrocytes Tak1 can phosphorylate the C-terminal residues of an R-Smad (Smad1) only as part of the Bmp signaling pathway. Moreover, we show that in neural crest cells, TGF-␤ stimulation is able to stimulate BMP R-Smad activation directly and that this process is, in part, mediated by Tak1. In fact, a recent study on established immortalized cell lines (of epithelial, endothelial, and mesenchymal origins) suggested that this unconventional TGF-␤-induced Bmp R-Smad activation is required for a subset of critical TGF-␤-induced cellular functions (5). Although our studies show that this unconventional R-Smad activation takes place in primary NC-derived craniofacial mesenchymal cells in FIGURE 5. Smad2 linker-region at Thr-220 is phosphorylated by nuclear kinases in palatal mesenchymal cells. A, representative immunoblot of primary palatal mesenchymal cells stimulated with TGF-␤2 in the presence of SB431542 (SB), 5Z-7-oxozeaenol (5Z), UO126 (UO1), and flavopiridol (Fla; 1 and 10 M (asterisk)). Protein lysates were analyzed for Smad2 C-terminal phosphorylation, Smad2 linker region phosphorylation (at Thr-220 and Ser-250), and Smad3 linker region phosphorylation (at Thr-179). The bar graphs show relative quantification of pSmad2-C, pSmad-L(T220), pSmad3-L(T179), and pSmad2-L(S250) (normalized to ␤-actin; n ϭ 3). Error bars, S.E. *, p Ͻ 0.05. B, representative immunoblot of primary palatal mesenchymal cells from Tgfbr1 FX/FX embryos were transduced with Ad-Gfp or Ad-Cre, the cells were stimulated with TGF-␤2, and protein lysates were analyzed for C-terminal Smad2 phosphorylation, p38 Mapk phosphorylation, and linker region phosphorylation at Smad2-L(T220) and at Smad3-L(T179). ␤-actin was loading control. C, palatal mesenchymal cells were stimulated by TGF-␤2 in the presence and absence of flavopiridol. Cytosolic and nuclear lysates were analyzed for Smad2 linker region phosphorylation at Thr-220 (pSmad2-L(T220)) and for Smad3 linker region phosphorylation at Thr-179 (pSmad3-L(T179)). Antibodies against PSMA2 and histone H3 were used to assess the purity of cytosolic and nuclear fractions (middle and bottom panels). D, schematic representation of Smad2 depicting Thr/Ser residues phosphorylated by nuclear Cdks and Jnk/Erk in the linker region and by TGF␤RI (RI) at the C terminus examined in this study. E, schematic summary illustrating the role(s) of Tak1 in phosphorylation of Smad2 (left side of diagram) in which Smad2 C-terminal phosphorylation levels were greatly enhanced in response to TGF-␤ stimulation when Tak1 was present, as well as Tgfbr1. Cdk-dependent phosphorylation of Thr-220 occurs in the nucleus. Phosphorylation of Ser-250 is Mapk-dependent but also inhibited by Cdk inhibitor, which may be due to some dependence on the prior phosphorylation of Thr-220 in the model of Aragón et al. (9). Deletion of Tak1 results in reduced C-terminal phosphorylation, leading to a reduced nuclear accumulation of Smad2, and reduced linker region phosphorylation (right side of diagram), which would be expected to reduce effectiveness in gene regulation according to the model of Aragón et al. (9).

Tak1 Signaling in the Craniofacial Neural Crest
vitro, it remains to be shown whether it plays a role in craniofacial ectomesenchyme during embryogenesis in vivo.
Linker regions in both TGF-␤ and BMP R-Smads are Ser/ Thr-rich and are known to be phosphorylated by several different kinases, e.g., GSK, Mapks, and Cdks, and it has been suggested that these post-translational modifications have both activatory and inhibitory regulatory functions in TGF-␤ signaling (38). Here we show that palatal and pharyngeal arch mesenchymal cells deficient in Tak1 show reduced agonist-induced linker region phosphorylation both at Thr-220 and Ser-250 in Smad2 in vitro and that particularly Thr-220 phosphorylation in the Smad2 linker region is reduced in neural crest cell-specific Tak1 mutants in vivo. Moreover, our results suggest that in NC-derived mesenchymal cells, Thr-220 and Ser-250 are phosphorylated by Cdks and Mapks, respectively. These findings differ from those made in fibroblast and melanoma cell lines, which showed that Mapks rather than Cdks are the primary kinases responsible for the linker region phosphorylation (46,47). Therefore, there seems to be distinct cell type-specific differences between the potency of particular kinases phosphorylate the specific residues.
It was recently suggested that ALP at Smad2-L(T220) plays a role in the so-called "action-turnover switch" function of R-Smads and may be required both for maximal TGF-␤-induced transcriptional activity and efficient termination of the signal in immortalized cell lines (9,10). Our present results show that, in primary craniofacial mesenchymal cells, Tak1, by functioning as an amplifier of C-terminal R-Smad phosphorylation, could have indirectly regulated ALP at Smad Thr220 because deficient Tak1 activity would result in reduced ligandinduced R-Smad activation and hence decreased R-Smad nuclear accumulation and less linker-region phosphorylation at Thr-220 by nuclear Cdks (Fig. 5E). However, our results do not exclude the possibility that Tak1 could also contribute to linker region phosphorylation by other mechanisms, such as regulating activity of other downstream kinases (Fig. 5E). Endogenous levels of Smad2-C phosphorylation and Smad Thr-220 phosphorylation were also reduced in Tak1/Wnt1Cre mutant palatal shelves, implying that Tak1 may modulate TGF-␤ superfamily signaling in vivo in the same way.
Interdependence of TGF␤RI and Tak1 Signaling?-Two recent studies addressed the mechanism of TGF-␤-induced Tak1 activation in HEK 293 cells (11,12). Although these studies agree that a ligand-induced interaction between TGF␤RI and Traf6 results in Traf6 autoubiquitination and subsequent Tak1 activation, they did not agree on a role for TGF␤RI kinase activity in TGF-␤-induced Tak1 activation. Here we show that, in craniofacial mesenchymal cells, the presence of TGF␤RI or its kinase activity is sufficient for Tak1-mediated signaling events downstream of TGF-␤2. To explore whether all the Tak1-dependent TGF-␤ transcriptional responses are dependent on the TGF␤RI kinase activity, we performed a genomewide transcriptomic analysis on prefusion palatal mesenchymal cells stimulated with TGF-␤2 in the presence or absence of either TGF␤RI or Tak1 kinase inhibitors. This assay demonstrated that many of the TGF-␤-induced transcriptional responses (both stimulatory and repressive) are simultaneously dependent on both TGF␤RI kinase and Tak1 kinase activities and that surprisingly few genes, e.g., Mllt3, respond to TGF-␤2 stimulation if TGF␤RI kinase activity is inhibited.
In conclusion, our combined evidence implies that Tak1 has a novel multimodal role in the craniofacial neural crest-derived mesenchyme in regulating activation of both TGF-␤-and Bmpinduced Smad-dependent and Smad-independent signaling processes. Tak1 deficiency in NC-derived mesenchymal cells leads to attenuation of both canonical and noncanonical TGF-␤ and Bmp signaling, which contribute to relatively subtle but consistent growth distortions in craniofacial structures. This discoordination in craniofacial growth likely results in delayed elevation of palatal shelves, which will never form a contact in the midline, resulting in cleft secondary palate.