Neural crest-specific deletion of Rbfox2 in mice leads to craniofacial abnormalities including cleft palate

Alternative splicing (AS) creates proteomic diversity from a limited size genome by generating numerous transcripts from a single protein-coding gene. Tissue-specific regulators of AS are essential components of the gene regulatory network, required for normal cellular function, tissue patterning, and embryonic development. However, their cell-autonomous function in neural crest development has not been explored. Here, we demonstrate that splicing factor Rbfox2 is expressed in the neural crest cells (NCCs), and deletion of Rbfox2 in NCCs leads to cleft palate and defects in craniofacial bone development. RNA-Seq analysis revealed that Rbfox2 regulates splicing and expression of numerous genes essential for neural crest/craniofacial development. We demonstrate that Rbfox2-TGF-β-Tak1 signaling axis is deregulated by Rbfox2 deletion. Furthermore, restoration of TGF-β signaling by Tak1 overexpression can rescue the proliferation defect seen in Rbfox2 mutants. We also identified a positive feedback loop in which TGF-β signaling promotes expression of Rbfox2 in NCCs.


Introduction
Approximately 3-4% of infants are born with congenital diseases. Collectively, craniofacial and cardiovascular abnormalities are the most common defects, contributing to more than one-third of the congenital diseases. Proper formation of these structures involves intricate processes such as proliferation, migration, and differentiation of NCCs, as well as their interaction with neighboring cells (Martik and Bronner, 2017;Plein et al., 2015). NCCs are a transient population of migratory progenitor cells that reside in the dorsal side of the neural tube. The NCCs separate from their neighboring cells through a delamination process and migrate via different pathways throughout the body to diverse locations, differentiating into multiple cell types at their respective destinations (Martik and Bronner, 2017). NCCs can be subdivided into five axial populations: cranial, cardiac, vagal, trunk and sacral NC cells. Lineage-tracing experiments in mouse embryos have demonstrated that NCCs differentiate into a diverse array of cell types including the craniofacial skeletal elements, autonomous nervous system of the heart and GI-tract, as well as smooth muscle cells of the cardiac OFT (Chai et al., 2000;Jiang et al., 2002;Jiang et al., 2000;Lee et al., 2004;Martik and Bronner, 2017;Plein et al., 2015;Waldo et al., 1999;Yoshida et al., 2008).
Cranial NCCs migrate from anterior portions of the folded neural tube and contribute to the formation of the skull, cartilage, and connective tissue, where they populate the first and second pharyngeal arches that give rise to cranial ganglia, the maxilla, the mandible, palates and other structures of the developing head (Chai et al., 2000;Jiang et al., 2002;Martik and Bronner, 2017). In mammals, both the primary and secondary palates are morphologically visible as early as E11.5 as demonstrated by an outgrowth from the oral side of the medial nasal and maxillary processes, respectively. The primary palate is formed by fusion of the maxillary prominence with the frontonasal prominence (Bush and Jiang, 2012;Dixon et al., 2011). Secondary palates are formed from the outgrowths of neural crest-derived mesenchyme that lie on either side of the developing tongue (Bush and Jiang, 2012). Initially, palatal shelves grow vertically flanking the developing tongue, but they elevate between E13.5 and E14.5 to a horizontal position above the tongue, grow toward the midline and fuse with each other to form an intact palate (Bush and Jiang, 2012). Functional defects in NCCs result in craniofacial malformations including cleft lip and/or cleft palate. Many transcription factors, chromatin remodeling factors, non-coding RNA and signaling molecules have been implicated in impaired neural crest development that result in cardio-craniofacial syndromes (Bélanger et al., 2018;Martik and Bronner, 2017;Plein et al., 2015;Singh et al., 2013;Strobl-Mazzulla et al., 2012). However, the cell-autonomous role of splicing regulators in neural crest biology remains unclear and warrants further investigation.
AS of pre-messenger RNAs is essential for regulating gene expression and creating proteomic diversity. Changes in exon inclusion or exclusion can produce multiple mRNAs and protein isoforms with related or distinct functions. The RNA-binding fox-1 (Rbfox) homolog proteins: Rbfox1, Rbfox2, and Rbfox3 are evolutionarily conserved splicing factors that have been implicated in diverse cellular processes such as cell proliferation, cell death and epithelial-mesenchymal transition (Kuroyanagi, 2009). A conserved RNA-Recognition Motif that recognizes UGCAUG element in the introns, flanking target exons, characterizes Rbfox proteins (Jin et al., 2003). Rbfox proteins regulate splicing either positively or negatively, depending on their binding sites. They promote exon skipping or eLife digest Abnormalities affecting the head and face -such as a cleft lip or palate -are among the most common of all birth defects. These tissues normally develop from cells in the embryo known as the neural crest cells, and specifically a subset of these cells called the cranial neural crest cells. Most cases of cleft lip or palate are linked back to genes that affect the biology of this group of cells.
The list of genes implicated in the impaired development of cranial neural crest cells code for proteins with a wide range of different activities. Some encode transcription factors -proteins that switch genes on or off. Others code for chromatin remodeling factors, which control how the DNA is packed inside cells. However, the role of another group of proteins -the splicing factors -remains unclear and warrants further investigation.
When a gene is switched on its genetic code is first copied into a short-lived molecule called a transcript. These transcripts are then edited to form templates to build proteins. Splicing is one way that a transcript can be edited, which involves different pieces of the transcript being cut out and the remaining pieces being pasted together to form alternative versions of the final template. Splicing factors control this process.
Cibi et al. now show that neural crest cells from mice make a splicing factor called Rbfox2 and that deleting this gene for this protein from only these cells leads to mice with a cleft palate and defects in the bones of their head and face.
Further analysis helped to identify the transcripts that are spliced by Rbfox2, and the effects that these splicing events have on gene activity in mouse tissues that develop from cranial neural crest cells. Cibi et al. went on to find a signaling pathway that was impaired in the mutant cells that lacked Rbfox2. Forcing the mutant cells to over-produce one of the proteins involved in this signaling pathway (a protein named Tak1) was enough to compensate for the some of the defects caused by a lack of Rbfox2, suggesting it acts downstream of the splicing regulator.
Lastly, Cibi et al. showed that another protein in this signaling pathway, called TGF-b, acted to increase how much Rbfox2 was made by neural crest cells. Together these findings may be relevant in human disease studies, given that altered TGF-b signaling is a common feature in many birth defects seen in humans. inclusion when they bind to either the upstream or downstream of the alternative exon (Jin et al., 2003;Ponthier et al., 2006). In addition to splicing, Rbfox proteins also regulate transcriptional gene networks (Fogel et al., 2012). Although all Rbfox proteins bind to the same recognition sequence, their in vivo functions are only partially redundant due to their markedly distinct expression pattern. Both Rbfox1 (A2BP1) and Rbfox2 (RBM9) are expressed in the neurons, heart and skeletal muscle (Gehman et al., 2012;Gehman et al., 2011;Singh et al., 2014;Wei et al., 2015). Unlike the aforementioned, Rbfox3 (NeuN or HRNBP3) expression is more restricted to neuronal tissues (Kim et al., 2014). Several diseases have been associated with dysregulation of splicing (Garcia-Blanco et al., 2004). For example, genetic deletion of Rbfox1 and Rbfox2 in the nervous system increased susceptibility to seizures and impaired cerebellum development, respectively (Gehman et al., 2012;Gehman et al., 2011). Morpholino-mediated knockdown of Rbfox1 and Rbfox2 in zebrafish embryos impaired cardiac and skeletal muscle functions (Gallagher et al., 2011). During skeletal muscle development, Rbfox2-mediated splicing is required for myoblast fusion and differentiation (Runfola et al., 2015;Singh et al., 2014). Decreased Rbfox2 expression has been reported in response to transverse aortic constriction in the mouse heart, and cardiac-specific deletion leads to pressure overload-induced heart failure (Wei et al., 2015). Abnormal expression of Rbfox2 has also been associated with hypoplastic left heart syndrome . Nutter et al. (2016) demonstrated that elevated Rbfox2 protein expression in diabetic hearts affects diabetes-induced AS of cardiac-specific genes. Despite extensive studies of Rbfox2 regulation on neuronal, cardiac and skeletal muscle cells, the cell-autonomous role of Rbfox2 in the functioning of neural crest cells has not been explored.
In the present study, we demonstrate that Rbfox2 is expressed in the neural crest cells (NCCs), neural crest-derived palate shelves, dorsal root ganglia, and somites. To determine the role of Rbfox2 during embryonic development, we deleted Rbfox2 using floxed Rbfox2 (Rbfox2 flox/flox ) and Pax3 Cre/+ knock in mice (Engleka et al., 2005;Gehman et al., 2012). We observed that Pax3 Cre/+mediated deletion of Rbfox2 results in neonatal lethality. Rbfox2 mutant (Rbfox2 Pax3-CKO ) embryos develop a cleft palate and defects in the development of craniofacial skeleton, suggesting defective cranial neural crest development. To determine whether cleft palate and craniofacial defects were due to deletion of Rbfox2 in the neural crest cells, we generated a neural crest-specific Rbfox2 mutant (Rbfox2 Wnt1-CKO ) using Wnt1 Cre/+ allele (Gehman et al., 2012;Lewis et al., 2013). Similar to Rbfox2 Pax3-CKO embryos, Rbfox2 Wnt1-CKO mutants developed a cleft palate and craniofacial skeleton defects and died postnatally at day 1. RNA-Seq analysis revealed an essential role of Rbfox2 in regulating splicing and expression of genes required for the development of cranial neural crest-derived structures. We demonstrate that Rbfox2-TGF-b-Tak1 signaling axis is impaired in the neural crestderived cells of the Rbfox2 mutant embryos and restoration of Tak1 expression in cultured palatal mesenchymal cells can rescue the proliferation defect observed in Rbfox2 mutants. We also identified a positive feedback loop by which TGF-b signaling components, Smad2/3/4, bind directly to the Rbfox2 promoter and regulate its expression in the neural crest cells. Together, these results reveal a highly regulated Rbfox2-dependent splicing and transcriptional program that modulates cranial neural crest development.

Rbfox2 is expressed in NCCs during mouse embryogenesis
To determine the expression pattern of Rbfox2, we performed Rbfox2 immunostaining on transverse sections from E9.5 to E11.5 embryos at different rostrocaudal axis. At E9.5, Rbfox2 is expressed in the premigratory NCCs at the dorsal neural tube, as well as in the migratory NCCs ( Figure 1A-C,F-H,K-M) throughout the rostrocaudal axis, with strong expression of Rbfox2 observed in the somites. Rbfox2 expression is gradually reduced in migratory NCCs ( Figure 1K-M). Rbfox2 expression was detected in the neural crest-derived craniofacial tissues including the palate shelves but not in the cardiac tissues such as OFT ( Figure 1D-E,I-J,N-O). To compare Rbfox2 expression with a known neural crest cell marker, we performed Rbfox2 and Pax3 immunostaining on transverse sections from E9.5, E10.5 and E11.5 embryos ( Figure 1P-U). At E9.5, Rbfox2 expression is identical to Pax3 in NCCs, dorsal neural tube, and somites ( Figure 1P and S). At E10.5, in addition to premigratory NCCs, Rbfox2 is also expressed in the dorsal root ganglia ( Figure 1Q). In contrast to Pax3 expression in the dorsal neural tube, the Rbfox2 expression is more restricted to the ventral neural tube ( Figure 1Q and T). At E11.5, Rbfox2 is expressed in the ventral neural tube and dorsal root ganglia ( Figure 1R). However, Pax3 is more restricted to the dorsal neural tube ( Figure 1U). At an early stage, Rbfox2 expression was similar to that of Pax3, which is transiently expressed in all premigratory, migratory NCCs and somites. In contrast to Pax3 expression in the dorsal neural tube, ! Figure 1. Rbfox2 is expressed in the neural crest cells during mouse embryonic development. Immunostaining for Rbfox2 was performed on E9.5, E10.5, E11.5 and E12.5 transverse sections at different rostrocaudal axis (A-R). Magnified view of the neural tube shows Rbfox2 expression in the premigratory (white arrows) and migratory (red arrows) neural crest cells (K-M). Rbfox2 expression in neural crest-derived tissues such as OFT and palate shelves (D-E, I-J, and N-O). Non-specific autofluorescence due to blood cells is observed in DA, MDA, OFT, RA and LA. Immunostaining for Rbfox2 and Pax3 was performed on adjacent sections from E9.5, E10.5, and E11.5 mouse embryos (P-U). Rbfox2 is expressed in the premigratory neural crest cells (white arrows) of the dorsal neural tube as well as in the migratory neural crest cells (red arrows). Nuclei were visualized by DAPI staining (blue). CAC, common atrial chamber and CV, cardinal vein; DGR, dorsal root ganglion; DM, dorsal mesocardium; DA, dorsal aorta; FV, fourth ventricle; LA, left atrium; MDA, midline dorsal aorta; NT, neural tube; OFT, outflow tract; PS, palatal shelves; RA, right atrium; S, somite; T, Tongue; TG, trigeminal ganglion; UV, umbilical vein; V, ventricle. Scale bars are 100 mm respectively. DOI: https://doi.org/10.7554/eLife.45418.003 Rbfox2 expression was restricted to the ventral neural tube at later stages. These results indicate that Rbfox2 is expressed in NCCs, dorsal neural tube, palate shelves, dorsal root ganglia, and somites.

Loss of Rbfox2 in NCCs results in cleft palate
In Rbfox2 Pax3-CKO embryos, development of primary palates is not affected. However, all of the Rbfox2 Pax3-CKO pups died neonatally, exhibiting a secondary cleft palate defect. To determine when the cleft palate defects were first evident, we examined Rbfox2 Pax3-CKO embryos at progressively earlier stages by analyzing their morphological and histological data (Figure 2A-L). At E12.5, the palatal shelves in both control and Rbfox2 Pax3-CKO embryos were initiated normally, growing vertically flanking the developing tongue (Figure 2A-B,G-H). No obvious morphological or histological differences were observed at this embryonic stage. At E15.5 and E18.5, the control palatal shelves were elevated above the tongue to a horizontal position and met each other at the midline along the anterior-posterior axis ( Figure 2C-F). However, in Rbfox2 Pax3-CKO embryos, palatal shelves were elevated above the tongue to a horizontal position, but they completely failed to fuse at the midline throughout the anterior-posterior axis ( Figure 2I-L). Since Pax3 Cre is expressed in both the neural crest and mesodermal derivatives, it was not clear if the cleft palate defects were due to Rbfox2 deletion in the neural crest cells. To determine the neural crest-specific requirement of Rbfox2, we generated neural crest-specific Rbfox2 mutant (Rbfox2 Wnt1-CKO ) embryos using Wnt1 Cre/+ mice (Gehman et al., 2012;Lewis et al., 2013). Similar to Rbfox2 Pax3-CKO , Rbfox2 Wnt1-CKO mice were born at the expected Mendelian ratio but died at P1 (Figure 2-figure supplement 1P-R). To determine the craniofacial defects, we analyzed Rbfox2 Wnt1-CKO palates at different time points. Similar to Rbfox2 Pax3-CKO embryos, palatal shelves in Rbfox2 Wnt1-CKO embryos were elevated above the tongue to a horizontal position, but completely failed to fuse at the midline throughout the anteriorposterior axis ( Figure 2M-T).
To determine the cellular mechanisms responsible for the impaired palatal growth in Rbfox2 mutant embryos, we performed cell proliferation, and apoptosis assays on transverse palatal sections from control and Rbfox2 mutant embryos. Ki67 immunohistochemistry revealed a significant reduction in cell proliferation in Rbfox2 mutants when compared with control embryos at E12.5, in which the morphological or histological changes were not evident ( Figure 3A-C). A significant difference in cell proliferation was more evident at E15.5 ( Figure 3D-F and G-I). TUNEL assay demonstrated no differences in cell death between the control and Rbfox2 mutants at E15.5 ( Figure 3J-K). E-cadherin expression is observed in the nasal, palatal and tongue epithelium and not affected by Rbfox2 deletion ( Figure 3L-M). To determine whether cleft palate defect was due to impaired neural crest cell migration, we performed lineage-tracing analysis at E15.5 in both control (Wnt1 Cre/+ : Rbfox2 flox/+ :R26 mTmG/+ ) and mutant (Wnt1 Cre/+ :Rbfox2 flox/flox :R26 mTmG/+ ) embryos. Labeled neural crest cells marked by GFP immunostaining were abundantly present in the palate shelves of mutant embryos ( Figure 3N

Craniofacial and axial skeletal defects in Rbfox2 mutants
To examine the nature and severity of the skeleton defects, we performed Alizarin Red S and Alcian Blue staining on Rbfox2 Pax3-CKO and Rbfox2 Wnt1-CKO embryos ( Figure 4). In both Rbfox2 Pax3-CKO and Rbfox2 Wnt1-CKO embryos, neural crest-derived bones such as frontal bones of the calvaria were hypoplastic and widely separated leaving a wide dorsal opening ( Figure 4C and M). Decreased ossification of nasal bone was observed in Rbfox2 Pax3-CKO mutant embryos compared to the controls ( Figure 4B-C and L-M). Reduction in the lower jaw or mandible size was observed in Rbfox2 mutant embryos ( Figure 4E-H and N-Q). Further analysis revealed that both the shape and size of most neural crest-derived bones including alisphenoid, premaxilla, palatal process of premaxilla, palatal process of maxilla and palatine are affected in Rbfox2 mutant embryos ( Figure 4J-K and R-S). The palatal process of palatine bone is also missing in Rbfox2 mutant embryos ( Figure 4K and S).
The whole embryo skeletal preparations displayed severe defects in the axial skeleton of Rbfox2-Pax3-CKO embryos (  To determine the effect of Rbfox2 deletion on neurons that populate the dorsal root, sympathetic and enteric ganglia, we performed whole mount neurofilament (2H3) staining to mark the differentiated neurons. The neurofilament 2H3 staining in Rbfox2 mutant embryos revealed abnormalities in the oculomotor (III), trochlear (IV) and hypoglossal nerve (XII cranial nerve) (Figure 4-figure supplement 3A). Oculomotor and trochlear nerves appear to be deformed. Analysis of cranial ganglia at higher magnification revealed that the roots of the hypoglossal cranial nerve are not fully developed. The hypoglossal cranial nerve is disorganized and shorter in Rbfox2 Pax3-CKO embryos (Figure 4-figure supplement 3A). Hypoglossal nerve defects could be secondary to defects in the hypoglossal cord which is derived from the occipital somites, and where Pax3 is expressed (Bajard et al., 2006).

Splicing and transcriptional changes in cranial neural crest-derived structures in response to Rbfox2 deletion
To determine the splicing program and transcriptional network regulated by Rbfox2 in vivo, we performed RNA-Seq profiling using poly(A) + RNA isolated from microdissected craniofacial tissues from E12.5 control and Rbfox2 mutant embryos ( Figure 5A). We performed RNA-Seq analysis at E12.5 because of the minimal morphological and structural changes observed in Rbfox2 mutant embryos at this stage. Multiplexed libraries were prepared for all replicates and sequenced together using Illumina HiSeq 4000 platform to produce 65-80 million, 151-nucleotide paired-end reads per sample (see Materials and methods for detail). Paired-end fastq sequence reads from each sample were aligned to mouse reference genome using ultrafast RNA-Seq aligner STAR with 82% average mapping rate and negligible ribosomal RNA contamination (<1%). Differential expression of genes and transcript isoforms between controls and Rbfox2 mutant samples were determined using two tools: MISO (Mixture of Isoforms), which quantitates the expression level of alternatively spliced genes and identifies differentially regulated isoforms or exons across samples and Cuffdiff, which finds significant changes in transcript expression, splicing, and promoter use. Using MISO analysis, we identified 81 differentially expressed transcripts from 59 genes in Rbfox2 mutant samples as compared with control ( Figure 5B and Figure 5-source data 1). However, Cuffdiff analysis identified 33 alternatively spliced transcripts from 30 genes in Rbfox2 mutant samples as compared with controls ( Figure 5C and Figure 5-source data 2).
Pathway enrichment analysis identified significant enrichment of genes that control cellular and anatomical structure morphogenesis ( Figure 5D). We analyzed the location of UGCAUG sequences in different gene features (5'-UTR, promoter, exons, introns, 3-UTR) of Rbfox2 target genes identified through Cuffdiff and MISO analysis. We performed an analysis of motif enrichment (AME) through the MEME suite (version 5.0.5) and identified significant UGCAUG motif enrichment only in the introns of the Rbfox2 target genes (adjusted p-value=1.4E-08) ( Figure 5E). We also performed AME on 75 randomly selected genes whose expression levels were comparable to the Rbfox2-target genes, but did not show any transcript-level differential expression in response to Rbfox2 deletion. However, no significant motif enrichment was observed (adjusted p-value=1.0E-01). Next, we used the FIMO tool in MEME to analyze the location of intronic UGCAUG sequences in the Rbfox2 target genes. Most UGCAUG sequences were located within 1000 bases of the exon on either flanking introns ( Figure 5F). Venn analysis was performed to identify a small set of high-confidence  alternatively spliced transcripts that could be validated in downstream experiments. Venn analysis identified 11 transcripts that were identified by both programs and constituted a high confidence set of differentially expressed and AS isoforms in Rbfox2 mutants ( Figure 5G-H and Figure 5-figure supplement 1). Specific genes identified by Venn analysis were selected for validation by RT-PCR based on their established function in the neural crest or craniofacial development. Using RNA isolated from control and Rbfox2 knockout craniofacial tissues, we validated the splicing changes identified by RNA-Seq by performing reverse transcriptase PCR (RT-PCR) in a selected group of genes such as mitogen-activated protein kinase kinase kinase 7 (Map3k7), fibronectin 1 (Fn1), periostin (Postn) and UAP1 UDP-N-acetylglucosamine pyrophosphorylase 1 (Uap1). Rbfox2 deletion significantly impacted splicing and expression of these candidate genes ( Figure 5I-L). For example, Rbfox2 deletion reduced the expression of the predominantly expressed Map3k7 transcript. The short Fn1 transcript was not present in Rbfox2 mutants. To determine that Rbfox2 directly bind to these target genes in vivo, we performed RNA immunoprecipitation (RIP) assays using anti-Rbfox2 antibody on palatal mesenchymal cells, followed by quantitative RT-PCR. We observed significant enrichment suggesting that Rbfox2 can directly bind to the RNA of these target genes and modulate splicing ( Figure 5J). To determine the transcriptional changes associated with the craniofacial defects, we analyzed genes that are differentially expressed between control and Rbfox2 knockouts. We identified 56 differentially expressed genes ( Figure 5K). We further validated the expression of candidate genes that were either differentially expressed or spliced by quantitative PCR in both control and Rbfox2 knockout tissues ( Figure 5L). We observed a significant reduction in Map3k7, Fn1, Myl1, Sfrs18 and Smarca2 in Rbfox2 knockout tissues ( Figure 5L). Together, we identified over 100 AS transcripts and 56 genes that are differentially expressed between control and Rbfox2 mutant embryos.
Impaired Rbfox2-TGFb-Tak1 signaling axis in Rbfox2 mutant embryos By analyzing alternatively spliced and differentially expressed genes from the RNA-Seq data, we observed that a number of genes known to affect TGF-b signaling pathway, such as Map3k7, Fn1, Meg3 etc. were significantly reduced in Rbfox2 mutants. This led us to hypothesize that TGF-b signaling pathway may be affected by Rbfox2 deletion. TGF-b signaling pathway involves both canonical and non-canonical signaling cascades. Recent work has shown that Tak1, encoded by Map3k7 is required for activation of both canonical and non-canonical signaling. To identify if TGF-b signaling pathway is affected by Rbfox2 deletion in vivo, we harvested palatal shelves from control and Rbfox2 mutant embryos and performed western blot analysis for Tak1 and phosphorylated Tak1. We found that both Tak1 and phosphorylated Tak1 levels were significantly reduced in Rbfox2 mutant embryos, demonstrating impaired TGF-b signaling pathway in neural crest-derived palate shelves ( Figure 6A). We further investigated the downstream signaling targets of Tak1 such as p38 Mapk, phosphorylated p38 Mapk, Smad2, and phosphorylated Smad2. The level of C-terminal Smad2 phosphorylation was significantly reduced; however, no change in total Smad2 was observed. Similarly, the level of phosphorylated p38 Mapk was significantly reduced, although there was no change in total p38 Mapk ( Figure 6A and Figure 6-figure supplement 1). Consistent with the reduction in its mRNA levels, Fn1 protein levels were also significantly reduced in Rbfox2 mutant neural crest as compared with controls ( Figure 6B and Figure 6-figure supplement 1). Next, we tested how neural crest-derived palatal mesenchymal cells respond to TGF-b stimulation. We established culture conditions to grow palatal mesenchymal cells from control and Rbfox2 mutant embryos. The neural crest cell's origin and purity were confirmed by growing cultures of Pax3 Cre/+ ;Rbfox2 flox/+ ; Rosa26 mTmG/+ embryos. Majority of the cultured cells were GFP positive confirming their neural crest origin ( Figure 6C). Consistent with the in vivo data, palatal mesenchymal cells isolated from Rbfox2 mutant embryos showed reduced levels of both Tak1 and phosphorylated Tak1 when compared with control following TGF-b stimulation ( Figure 6D). In addition, no significant change in pTak1/ Tak1 ratio was observed, suggesting that Tak1 activation was not affected by Rbfox2 deletion ( Figure 6E). Similarly, TGF-b-induced C-terminal Smad2 and p38 Mapk phosphorylation levels were reduced in Rbfox2 mutant cells when compared with controls, while no significant change was observed in total Smad2 or p38 Mapk ( Figure 6D-E and Figure 6-figure supplement 1). Quantification of these proteins showed a significant reduction in the ratio of pSmad2-C/Smad2-C and p-p38 Mapk/p38 Mapk, suggesting that Rbfox2-TGF-b-Tak1 signaling axis was impaired in Rbfox2 mutant cells ( Figure 6D-E and Figure 6-figure supplement 1).
To determine whether Tak1 overexpression can restore TGF-b signaling pathway and rescue the proliferation defects seen in Rbfox2 mutant embryos in vivo, we cultured palatal mesenchymal cells from control and Rbfox2 mutant embryos. We then transfected Rbfox2 mutant cells with either empty plasmid (pcDNA3) or plasmid expressing Tak1 (pcDNA3-Tak1), and performed Ki67 immunostaining ( Figure 6F-H). We confirmed that both Tak1 and phosphorylated Tak1 levels were significantly increased after Tak1 transfection in Rbfox2 mutant cells ( Figure 6G). Consistent with the in vivo data, Rbfox2 mutant palatal cells proliferate slower than control cells and Tak1 overexpression can rescue the proliferation defects ( Figure 6H). To determine the effect of Tak1 overexpression on the expression of Rbfox2-dependent genes, we performed qRT-PCR for Map3k7, Fn1, Myl1, Sfrs18 and Smarca2 on control and Rbfox2 mutant cells transfected with either empty plasmid vector or plasmid expressing Tak1 ( Figure 6-figure supplement 2). Compared with empty vector transfected controls, we observed significant increase in the expression of Map3k7 and Fn1 in Tak1 overexpressing Rbfox2 mutant cells. No change in Myl1, Sfrs18 and Smarca2 expression was observed ( Figure 6-figure supplement 2). Together, these results demonstrate that Rbfox2 deficiency leads to impaired Rbfox2-TGF-b-Tak1 signaling axis, resulting in reduced palatal cell proliferation. In addition, Tak1 overexpression restores TGF-b signaling pathway and rescues the proliferation defects in Rbfox2 mutant cells.

Rbfox2 expression is regulated by TGF-b signaling pathway in a feedback loop
To determine whether TGF-b signaling pathway regulate Rbfox2 expression in a positive feedback loop, we analyzed Rbfox2 expression in wildtype palatal mesenchymal cells after TGF-b stimulation. We observed a significant increase in Rbfox2 expression after TGF-b treatment ( Figure 7A  Together, these results suggest that TGF-b induces Rbfox2 expression either through Tak1-dependent or -independent canonical pathways. To further investigate how Smad-dependent canonical pathway regulates Rbfox2 expression, we analyzed 2.5 kb Rbfox2 promoter and identified two sites with multiple Smad binding elements (SBEs). Rbfox2 promoter fragment (1.6 kb) with multiple SBEs was PCR-amplified, cloned into a luciferase reporter plasmid, and tested in luciferase reporter assays (Figure 7-figure supplement 2). Smad2, Smad3 or Smad4 significantly activated the Rbfox2 luciferase reporter in the presence/absence of recombinant TGF-b. However, in the presence of recombinant TGF-b, the fold activation was much higher when compared with no TGF-b stimulation ( Figure 7E-G). Using the wild-type palatal mesenchymal cells in the presence or absence of TGF-b, we next tested Smad2/3 binding activity to SBEs identified in the Rbfox2 promoter in vivo by chromatin immunoprecipitation (ChIP) assays. Our data indicate that Smad2/3 binds directly to these sites in vivo. Moreover, we observed enrichment in Smad2/3 chromatin binding after TGF-b treatment ( Figure 7H). Together, these results demonstrate that Rbfox2 expression in neural crest-derived palatal mesenchymal cells is tightly regulated by TGFb signaling pathway ( Figure 7I). Representative Rbfox2 western blot and quantification from primary palatal mesenchymal cells stimulated with recombinant TGFb with and without, Smad3 inhibitor (SIS3) or Tak1 inhibitor (5Z-7-Oxozeaenol) or p38 inhibitor (SB203580) at 6 and 24 hr time points. b-Actin was used as a loading control (B-D). *, p<0.05; **, p<0.01 (one-way ANOVA between groups, Bonferroni's multiple comparisons test). The Rbfox2-luciferase reporter was transfected in HEK293T cells with or without recombinant TGFb in the presence or absence of Smad2 (E) or Smad3 (F) or Smad4 (G). ChIP assay using chromatin from untreated or TGFb treated primary palatal mesenchymal cells and Smad2/3 antibody. Predicted binding sites in Rbfox2 promoter were tested (H). Model depicting the interaction between the TGF-b signaling pathway and alternative splicing factor Rbfox2 in neural crest cells. Rbfox2 regulates alternative splicing and transcription of neural crest genes. In a feedback loop, Rbfox2 expression is regulated by the canonical TGFb signaling pathway (red arrows) (I). DOI: https://doi.org/10.7554/eLife.45418.022 The following figure supplements are available for figure 7:

Discussion
During embryonic development, various organ formations require precise spatial and temporal regulation of gene expression. Traditionally, developmental studies were more focused on the role of transcription factors and signaling pathways. It is only in recent years that the importance of splicing factors has been demonstrated in regulating various developmental processes. The gene regulatory network required to orchestrate neural crest development is very well defined (Gammill and Bronner-Fraser, 2003;Hutson and Kirby, 2003;Sauka-Spengler and Bronner-Fraser, 2008;Simõ es-Costa et al., 2014). However, the cell-autonomous role of splicing factors in neural crest development is poorly investigated. In the present study, we demonstrate a critical role for splicing regulator Rbfox2 in NCCs. We show that Rbfox2 is expressed in pre-migratory and migratory NCCs, neural crest-derived palate shelves, dorsal root ganglia, and somites. Genetic deletion of Rbfox2 using Pax3 Cre/+ or Wnt1 Cre/+ mouse strains affected neural crest cells and their derivatives, causing severe craniofacial defects including cleft palate. We show that cleft palate defect was due to impaired palate cell proliferation and not due to cell death or impaired NCCs migration. To examine the effect of Rbfox2 deletion on cranial NCCs, we examined the cranial NC-derived craniofacial skeletons and found that majority of the NC-derived bones were affected in Rbfox2 mutants.
NCCs contribute to the formation and septation of the cardiac OFT, as well as patterning and remodeling of aortic arch arteries. In contrast to a fully penetrant cleft palate and craniofacial bone defects, no cardiac defects were observed in Rbfox2 mutants. This surprising finding led us to investigate if Rbfox2 is expressed during OFT development. We found Rbfox2 expression was not detected in the NC-derived cardiac tissues; thereby explaining the lack of OFT defects. NCCs also contribute to the peripheral tissues such as nervous systems, thymus, adrenal gland and others (Dupin and Sommer, 2012). With the exception of cranial nerves, all other neural crest-derived tissues analyzed develop grossly normal, suggesting that Rbfox2 may not be or only transiently expressed in these NC-derived tissues. These findings indicate that Rbfox2 is required for the development of a narrow subset of the NCCs.
Since Pax3 Cre is also active in non-neural crest-derived tissues such as somites and limb and diaphragm muscles (Bajard et al., 2006;Engleka et al., 2005), we included these tissues in our analysis. In contrast to the controls, ectopic bone formation and fusion of vertebral bodies was observed in Rbfox2 Pax3-CKO embryos, most likely the reason for the straight vertebral column. As the metameric organization of the axial skeleton is derived from the somites, these results demonstrate that Rbfox2 in the somites is necessary for proper development of the vertebral column. No obvious defects in the limb and diaphragm muscles were observed. Since both Rbfox1 and Rbfox2 are coexpressed in skeletal muscle, it is possible that Rbfox1 is able to compensate for the loss of Rbfox2, thus preventing any developmental defects. Recently, Singh et al. (2018) generated skeletal muscle-specific Rbfox1/2 double knockout and observed a severe reduction in muscle mass, suggesting functional redundancy.
Splicing is a tightly regulated process required for increasing the transcriptome complexity using a finite set of genes, (Baralle and Giudice, 2017;Revil et al., 2010). It enhances proteomic diversity by increasing the number of distinct mRNAs transcribed from a single gene. Depending upon the type of tissues and organs, subtle changes in mRNA by splicing may alter RNA stability and/or function, protein interaction networks by either removing or inserting protein domains, subcellular localization and gene expression (Baralle and Giudice, 2017;Garcia-Blanco et al., 2004;Revil et al., 2010). Genetic mutations in splicing regulators have been reported in various human diseases (Cieply and Carstens, 2015;Garcia-Blanco et al., 2004). Here, we not only demonstrate that Rbfox2 is essential for the development of tissues derived from NCCs, but also uncover over 100 Rbfox2-dependent splicing events that occur during neural crest development.
RNA sequencing analysis revealed that Rbfox2 deletion altered splicing and expression of genes involved in the neural crest or craniofacial development. For example, Map3k7 encodes transforming growth factor b (TGF-b)-activated kinase 1 (Tak1) required for the proliferation of palatal mesenchymal cells. Neural crest-specific deletion of Tak1 results in the cleft palate (Song et al., 2013;Yumoto et al., 2013). Fibronectin 1 (Fn1), which is a component of the extracellular matrix, has various alternatively spliced variants. Wang et al. recently demonstrated that neural crest-specific deletion of Fn1 leads to the cleft palate development (Wang and Astrof, 2016). Reduced expression of both Map3k7 and Fn1 was observed in Rbfox2 mutant tissues. We also identified a number of candidate genes that are novel and their role in neural crest/craniofacial development has not been established. In the present study, in addition to changes in splicing, we also identified 56 genes that are differentially expressed between control and Rbfox2 mutant embryos. A number of genes implicated in craniofacial bone development such as Igf1, Wnt5a, Fn1, and Aldh1a2 etc. were downregulated in Rbfox2 mutants (Halilagic et al., 2007;Lohnes et al., 1994;Yamaguchi et al., 1999;Zhang et al., 2002). We found that only five (Meg3, Ccnl2, Smarca2, Tpm1, and Fn1) of 56 differentially expressed genes were also alternatively spliced, suggesting that Rbfox2 regulates transcriptional gene networks apart from alternative splicing. This is not surprising considering Rbfox2 has been reported to regulate gene expression patterns by different mechanisms Fogel et al., 2012;Lee et al., 2016;Wei et al., 2016). For example, Rbfox2 can affect transcript stability by directly binding to the 3'UTR of target genes Lee et al., 2016). Rbfox2 may affect gene expression by recruiting polycomb complexes to the DNA (Wei et al., 2016). Future work in this direction will help to determine the mechanism by which Rbfox2 regulates expression of neural crest genes. Recent studies have defined the transcriptional network required for regulating many aspects of cranial neural crest biology (Simõ es-Costa et al., 2014). However, the mechanisms by which splicing factors could be integrated into these gene regulatory networks need to be further explored.
Recent studies in mice demonstrated that Tak1 is required in the NCCs to activate both TGF-binduced canonical (R-Smads) and non-canonical (p38 Mapk) pathways (Yumoto et al., 2013). Fn1, a positive regulator of TGF-b signaling, was downregulated at both mRNA and protein levels. Fn1 and its integrin receptor positively regulate TGF-b signaling by promoting the receptor complex formation on the cell surface (Tian et al., 2012). Fibronectin also regulates TGF-b by controlling the matrix assembly of latent TGF-b-binding protein-1 (Dallas et al., 2005). In a reciprocal manner, TGF-b1 induces Fn1 expression via a Smad independent pathway (Hocevar et al., 1999). Meg3 modulates the expression of TGF-b pathway genes by binding to the distal regulatory elements (Mondal et al., 2015). Similar to Rbfox2 mutants, neural crest-specific deletion of Map3k7 or Fn1 leads to craniofacial defects including cleft palate (Song et al., 2013;Wang and Astrof, 2016;Yumoto et al., 2013). Chondrocyte-specific deletion of Tak1 results in severe chondrodysplasia with impaired ossification and joint abnormalities including tarsal fusion (Shim et al., 2009). Osteoblast-specific deletion of Tak1 results in clavicular hypoplasia and delayed fontanelle fusion (Greenblatt et al., 2010). Interestingly, loss of Rbfox2 results in similar defects such as fused cervical bones, hypoplastic craniofacial bone, delayed ossification and fusion of cranial bones. Consistent with these findings, heterozygous mutations in MAP3K7 cause cardiospondylocarpofacial syndrome, as characterized by craniofacial and cardiac defects including dysmorphic facial bones and extensive posterior cervical vertebral synostosis (Le Goff et al., 2016;Wade et al., 2016). Altogether, the striking similarities in craniofacial and skeletal phenotypes between Tak1 and Rbfox2 mutants suggest that Tak1 is a downstream target of Rbfox2 and it may significantly contribute to the phenotype observed in Rbfox2 mutant embryos. Furthermore, restoration of TGF-b signaling by Tak1 overexpression can rescue the proliferation defects seen in Rbfox2 mutant embryos. This indicates that Tak1 regulates neural crestderived tissues downstream of Rbfox2. It is possible that other Rbfox2 target genes identified in our RNA-Seq screen may be responsible or contribute to the craniofacial phenotype present in Rbfox2 mutant embryos. Thus, further functional characterization and investigation of their expression and splicing patterns are clearly warranted.
Since a single splicing factor affects the expression/splicing of numerous genes and displays profound downstream effects, changes in the expression levels of splicing factors must be tightly regulated during embryonic development (Baralle and Giudice, 2017;Revil et al., 2010). We found that expression of Rbfox2 in cranial NCCs is dependent on TGF-b signaling. Furthermore, Rbfox2 is required for TGF-b-Tak1 signaling axis in embryonic neural crest development. Given that altered TGF-b signaling is well studied in multiple human congenital malformations and syndromes, our observation may be relevant in human disease studies. In summary, we have provided evidence that Rbfox2 modulates neural crest development.

Alcian blue/Alizarin red and Von Kossa staining
Alcian Blue/Alizarin Red staining of bone and cartilage was performed as described previously (Singh et al., 2005b). Briefly, euthanized embryos were placed in tap water for 1-2 hr at 4˚C. Embryos were placed in 65˚C water for 30 s allowing easy removal of skin. Visceral organs were removed under the microscope and the embryos were placed in 100% ethanol for 2-3 days at room temperature. Cartilage staining was performed using Alcian blue solution (150 mg/L Alcian blue 8GX in 80% ethanol/20% acetic acid) for 2-3 days. Embryos were rinsed and post-fixed overnight in 100% ethanol. Bone staining was performed using Alizarin red solution (50 mg/L Alizarin red S in 0.5% KOH) for 1-2 days at room temperature. Embryos were incubated in 0.5% KOH until most of the soft tissues were digested. The 0.5% KOH solution was replaced with 20% glycerol in water and incubated at room temperature until tissues cleared completely. Imaging was done using an inverted Olympus dissecting microscope.

RNA sequencing, data analysis and, RT-PCR validation
Craniofacial tissue was microdissected from E12.5 control and knockout embryos in cold PBS. Three independent biological replicates were used for each genotype group. Tissues were homogenized and RNA was isolated using a PureLink RNA Mini kit from Thermo Fisher (Cat. no. 12183018A). Sequencing libraries of poly(A)+RNA from 3 control and three mutant samples were prepared using the TruSeq Stranded mRNA Library Prep Kit (Illumina) according to manufacturer's instructions. Biological replicates were individually barcoded and pooled for paired-end sequencing using Illumina HiSeq4000 platform at the Genome Institute of Singapore. For each sample, approximately~60-80 million paired-end reads of 151 bp were used for genome-guided alignment. Paired-end fastq sequence reads from each sample were aligned to mouse reference genome (GRCm38) using ultrafast RNA-seq aligner STAR (Dobin et al., 2013) with 82% average mapping rate and negligible ribosomal RNA contamination (<1%). Differential expression of genes and transcripts between controls and knockout samples were determined using two tools: MISO (Mixture of Isoforms) (Katz et al., 2010) and Cuffdiff (Trapnell et al., 2013).

Luciferase assay
Luciferase assay was performed as previously described (Singh et al., 2016;Singh and Epstein, 2012). HEK293T cells are the most commonly used cell line for monitoring the activity of the TGF/ SMAD signaling pathway. Briefly, HEK293T cells were seeded in 12-well plates for 24 hr before transfection. The Rbfox2 luciferase reporter plasmid along with other indicated plasmids (Smad2, Smad3 or Smad4) was co-transfected using FuGENE6 reagent (Promega, catalog no. E2691). To normalize transfection, 50 ng of lacZ expression plasmid was also transfected together with other indicated plasmids. Cell extracts were prepared 60 hr post-transfection using lysis buffer (Promega, catalog No. E3971). Luciferase activities were assayed using Luciferase Reporter Assay System kit (Promega, catalog no. E1500). Lysates were also assayed for b-galactosidase activity using the b-Galactosidase Enzyme Assay System (Promega, Cat. no. E2000). Luciferase reporter activity was normalized to b-galactosidase activity. The luciferase assay results were reproduced in at least three independent experiments. All experiments were performed in duplicate, and the representative data are shown in the bar graphs.

ChIP and RNA immunoprecipitation assays
ChIP experiments were performed as previously described (Singh et al., 2016). ChIP assay was performed on either unstimulated or TGFb-stimulated palate mesenchymal cells using Smad2/3 antibody (Abcam, Cat. no. ab207447), according to Millipore Chip Assay Kit protocol with minor modifications (Catalog no #17-295). RNA-IP experiments were performed as previously described with minor modifications (Niranjanakumari et al., 2002).

Statistical analysis
Statistical analyses were performed using the two-tailed Student's t-test. Data are expressed as mean ± SD. Differences were considered significant when the p-value was <0.05. One-way analysis of variance (ANOVA) was used to assess statistical differences between groups. Significant ANOVA results were further analyzed by Bonferroni's multiple comparisons test (*, p<0.05; **, p<0.01; ***, p<0.001; NS, not significant).
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.