An FGF autocrine loop initiated in second heart field mesoderm regulates morphogenesis at the arterial pole of the heart

Many congenital heart defects are caused by abnormal remodeling of the arterial pole of the heart, called the outflow tract (OFT). Cells are added to both poles of the heart from a second population of cells located in pharyngeal and splanchnic mesoderm dorsal to the heart tube, called the second heart field (SHF), which is crucial for the formation of the OFT(Buckingham et al., 2005). After looping of the heart, the myocardium secretes a specialized extracellular matrix (ECM) called the cardiac jelly that forms internal cushions; these cushions ultimately form the septa and the valves. The cushion jelly is invaded by endothelial cells lining the OFT myocardium that undergo an endothelial-to-mesenchymal transformation (EMT), and by a subset of neural crest cells called the cardiac neural crest (CNC)(Kirby, 2006). These events require BMP, TGFβ, WNT and semaphorin signaling from the OFT myocardium,in addition to proper cardiac jelly composition(Armstrong and Bischoff, 2004; High and Epstein, 2007). Thus,dysfunction of many different cell types and signaling pathways can contribute to abnormal OFT morphogenesis.

At midgestation, the OFT is remodeled by spiraling and fusion of the cushions to septate the OFT into the aorta and pulmonary artery; this occurs concurrent with rotation and alignment of the OFT relative to the ventricles. When alignment/rotation of the septated OFT is disrupted, transposition of the great arteries (TGA) or double-outlet right ventricle (DORV) occurs. Complete failure of OFT septation results in persistent truncus arteriosus (PTA), in which the primitive OFT (the truncus arteriosus) is not divided into the aorta and pulmonary artery. These defects disrupt the partitioning of blood flow required for adequate oxygenation and are lethal in mice and humans.

The importance of FGF8 signaling for heart development is clear. Fgf8-null mutants die at gastrulation(Sun et al., 1999). However, Fgf8 hypomorphs survive to birth with OFT septation and alignment/rotation defects (Abu-Issa et al., 2002; Frank et al.,2002). Conditional mutagenesis has been used to determine the spatiotemporal requirements for Fgf8 function in different aspects of OFT morphogenesis (Ilagan et al.,2006; Macatee et al.,2003; Park et al.,2006). Mutation of Fgf8 in mesodermal heart precursors in the primitive streak can prevent OFT formation and cause embryonic death;however, survivors display OFT rotation/alignment defects. When Fgf8is inactivated later in development, in OFT precursors in the SHF and in pharyngeal endoderm, the mutants survive, but all have PTA(Park et al., 2006). Although Fgf10 is expressed in the SHF, and Fgf15 in multiple tissues in the pharyngeal arches, only Fgf15-null mutants have OFT defects(Marguerie et al., 2006; Vincentz et al., 2005).

We have made progress in identifying the sources and FGF ligands required for OFT development, but their cellular targets remain unknown. The accepted paradigm is that FGFs signal in a paracrine manner to target cells via the FGF receptor tyrosine kinases (RTKs, FGFR1-4). Mitogenic assays suggest that ligand-receptor preferences exist (Zhang,X. et al., 2006), but overlapping expression patterns and ablation studies reveal functional redundancy (Sun et al., 2002). Variations in ECM composition are likely to modify ligand-receptor interactions and signaling in a context-dependent manner. Thus, defining functionally relevant ligand-receptor interactions on target cells that regulate specific morphogenetic events in vivo is an important endeavor.

Some insight into these crucial in vivo interactions can be obtained by comparing phenotypes of FGF receptor and ligand mutants. Fgfr1-null mutants die at gastrulation (Deng et al.,1994), whereas Fgfr2-null mutants die post-implantation(Arman et al., 1998). Previous manipulations of Fgfr1/2 function that bypass lethality reveal crucial roles for these receptors in many processes and suggest that they are important for cardiovascular development(Marguerie et al., 2006; Moon et al., 2006; Trokovic et al., 2003). By contrast, Fgfr3 and Fgfr4 double-null mutant mice survive without cardiovascular defects (Weinstein et al., 1998).

Here, we conditionally inactivate Fgfr1 and Fgfr2 and conditionally overexpress sprouty 2 (Spry2, which encodes an FGF signaling antagonist) in different cell types to identify the direct cellular targets of FGF signals required for OFT remodeling. Our results reveal that although published evidence points to CNC and/or OFT endothelial cells as crucial paracrine targets (Kirby,2006; Presta et al.,2005), disrupting FGF signaling to these populations does not prevent OFT remodeling. Rather, we show that interrupting autocrine FGF signaling in SHF mesoderm causes an OFT myocardial secretory defect; this secondarily perturbs endothelial EMT and CNC invasion in association with altered BMP and TGFβ signaling. Graded alterations in OFT development with varying FGF receptor gene dosage reveal a marked sensitivity to FGF signaling in target cells. We conclude that the specialized secretory and signaling properties of the primitive OFT that drive OFT morphogenesis are regulated by an autocrine FGF signaling loop.

Construction and activity of the Fgfr1, Fgfr2 and Fgf8conditional alleles (denoted as `c' alleles throughout) were described previously (Park et al., 2006; Trokovic et al., 2003; Xu et al., 2002; Yu et al., 2003). Generation and characterization of Wnt1Cre, Ap2αIRESCre, Pax3Cre,Mesp1Cre, Isl1Cre, Tie2Cre (Tie2 is also known as Tek -Mouse Genome Informatics), Foxa2MCM and Spry2-GOF alleles have also been reported previously (Basson et al., 2008; Danielian et al., 1998; Engleka et al.,2005; Jiang et al.,2000; Kisanuki et al.,2001; Macatee et al.,2003; Park et al.,2006; Park et al.,2008; Saga et al.,2000). The Isl1Cre and Mesp1Cre males used to generate the mutants are in mixed backgrounds: Isl1Cre is 50% C57Bl6 and 25% each BlSw and Sv129; Mesp1Cre is 50% C57Bl6 and 25% each ICR and SV129. These males were bred to Fgfr1/2 conditional females that are in a complex mixed background (including C57Bl6, Sv129 and others). Thus,the embryos arising from each cross are outbred. To generate the Fgfr1;IRESCFP and Fgfr2;IRESYFP alleles in mice, we inserted IRES;FP;frt-flanked neomycin cassettes into the 3′ untranslated regions of each gene upstream of the polyadenylation signal. All experiments were in accordance with institutional and national guidelines and regulations.

Collagen gels were made as previously described(Camenisch et al., 2002). OFT segments were isolated, opened to expose endocardium, placed on the gel and allowed to attach for 12 hours without addition of culture media. Then 1×M199 containing 1% fetal bovine serum (FBS), 0.01%insulin-transferrin-selenium (ITS) and 1% PenStrep (Gibco-BRL) was added and the explants cultured for 48 hours. For co-cultures, Fgf8^c/c;Rosa26^lacZ/lacZ and Fgf8^c/c females were bred with Fgf8^+/-;Isl1Cre males to obtain labeled Fgf8;Isl1Cre mutants and controls. OFTs from mutants and controls were placed immediately adjacent to one another. Explants were fixed and stained using standard protocols.

YFP and CFP detection was performed using rabbit anti-GFP and Texas Red-conjugated anti-rabbit secondary antibodies (Molecular Probes) on cryosectioned specimens. Anti-phosphohistone H3 (Ser10, Cell Signaling Technology), anti-ISL1 (Developmental Studies Hybridoma Bank), anti-AP2α(TCFAP2α - Mouse Genome Informatics) (3B5, Developmental Studies Hybridoma Bank), anti-phosphorylated (p) ERK1/2 and SMAD1/5/8 (Cell Signaling Technology) and FITC-conjugated anti-mouse secondary (Molecular Probes)antibodies were used. In situ hybridization was performed with a standard protocol (Grove et al., 1998)and digoxigenin-labeled antisense RNA probes. Alkaline phosphatase andβ-galactosidase staining were performed using published protocols(Frank et al., 2007; Lobe et al., 1999).

E9.5 OFTs were dissected and stored in RLT buffer (Qiagen) at -80°C. Seven specimens of each genotype (Fgf8^c/+, control; Fgf8^c/-;Isl1Cre, mutant) were pooled to generate each sample. Total RNA was extracted from four samples (RNeasy Micro Kit, Qiagen). Agilent two-color LRILAK labeling, the Agilent two-color GE hybridization/wash protocol, and the Agilent 5-micron XDR scanning protocols were carried out by the University of Utah Microarray Core Facility. RNA (100 ng) was reverse-transcribed to cDNA using the SuperScript III First-Strand Synthesis System (Invitrogen). Quantitative RT-PCR was performed with iQ SYBR Green Supermix on the iCycler system (Bio-Rad). Hprt transcript was used as the reference level.

This experiment was run in quadruplicate on Agilent mouse whole-genome expression arrays. The array image data were quantitated using Agilent Feature Extraction software (version 9.5.1.1). Subtle intensity-dependent bias was corrected with LOWESS normalization, with no background subtraction. Statistical analysis of normalized log-transformed data was performed in Gene Sifter(www.genesifter.net). Differentially expressed transcripts were defined (adjusted for multiple testing using the Benjamini and Hochberg method) as P<0.05. Spots with an intensity below background were removed prior to statistical analysis. Data were hierarchically clustered with Spotfire (TIBCO) and heat maps for selected genes were generated. Log-transformed, normalized gene intensity data were clustered by the UPGMA method; Pearson's correlation was the distance metric.

Loss of Fgf8 function in the SHF and pharyngeal endoderm in Fgf8^c/-;Isl1Cre conditional mutants (hereafter referred to as Fgf8;Isl1Cre) causes PTA (Park et al., 2006). We found that the OFT cushions of these mutants contain markedly less cardiac jelly and fewer mesenchymal cells than controls(Fig. 1A-H; see Fig. S1 in the supplementary material). The mutant OFTs were short and aberrantly angulated.

To distinguish the sources of cushion mesenchymal cells at early stages of OFT septation, we used the conditional Rosa26^lacZ reporter(Soriano, 1999) and Tie2Cre (Kisanuki et al.,2001) to label endothelial cells (which constitute the endocardial lining of the OFT). Tie2Cre activity was uniform in the OFT endothelium by E8.25 (Fig. 1I-K). This experiment shows that most cells in the proximal cushions at E11.5 are of endothelial origin; although CNC cells have invaded the distal OFT, few have as yet migrated into the proximal cushions(Fig. 1E,G,J and data not shown). We explanted proximal cushions from E9.25 Rosa26^lacZ;Tie2Cre embryos onto collagen gels and found that the cells from the explant that undergo EMT to invade and migrate into the gel are β-galactosidase-positive(Fig. 1K,L), confirming their endothelial origin. Since the OFT cushions of Fgf8;Isl1Cre mutants were hypoplastic along their entire proximodistal extent (the proximal cushions were both thinner and had fewer cells, whereas the distal cushions were thinner but with no change in cell density)(Fig. 1; see Fig. S1 in the supplementary material), both endothelial and CNC cells are affected in these mutants.

In order to understand how FGF8 regulates endothelial and CNC invasion of the OFT cushions, we sought to identify the direct cellular targets of FGF8 and of other FGF ligands required for OFT remodeling. The overlapping phenotypes reported in different types of Fgf8, Fgfr1 and Fgfr2 mutant mice suggest that FGF8 signals through these receptors during embryogenesis. We first determined which cells within the pharyngeal arches and SHF express these receptors. We generated novel alleles to fluorescently label cells expressing Fgfr1 and Fgfr2 by targeting an IRES (internal ribosome entry site) and CFP to Fgfr1,and an IRES and YFP to Fgfr2. We detected ubiquitous expression of both receptors in the anterior embryo, including all cells in the pharynx at E8.5-9.5 (see Fig. S2 in the supplementary material), with variable levels of signal in different cell types.

Based on these expression data, we systematically ablated these receptors in the mesodermal precursors of the OFT in the SHF and in other pharyngeal tissues. We assayed whether ablation of Fgfr1 and/or Fgfr2in a tissue phenocopies the OFT septation defects seen in Fgf8;Isl1Cre mutants, or the OFT alignment defects seen in Fgf8;Mesp1Cre mesodermal mutants(Park et al., 2006). Each of the Cre drivers used has an onset of Cre-mediated recombination in the desired tissue that is prior to the requisite window of FGF8 signaling at E8.5 (8- to 10-somite stage) that we previously identified for OFT remodeling(Park et al., 2006).

In complementary experiments, we employed conditional Spry2gain-of-function (Spry2-GOF) to inhibit ERK phosphorylation downstream of activated RTK signaling, including that of FGFRs(Hanafusa et al., 2002). After Cre-mediated recombination, the Spry2-GOF transgene constitutively expresses Spry2 (Fig. 2) (Basson et al.,2008). Activation resulted in markedly increased production of Spry2 mRNA relative to controls(Fig. 2F,H), and effective antagonism of FGF signaling was evident by decreased ERK activation(Fig. 2J,J′) and decreased expression of the FGF8 target gene Erm (Etv5)(Fig. 2L)(Park et al., 2006; Raible and Brand, 2001; Roehl and Nusslein-Volhard,2001). Pea3 (Etv4) expression was also decreased at E8.5 (not shown).

To test whether CNC cells are direct targets of the FGF signaling required for OFT remodeling, we used the well-characterized Wnt1Cre and Ap2αIRESCre drivers to ablate Fgfr1 and Fgfr2 function in premigratory neural crest, and in premigratory neural crest and pharyngeal ectoderm, respectively(Jiang et al., 2000; Macatee et al., 2003). Surprisingly, ablation of either receptor, independently or in combination,did not disrupt OFT remodeling (see Table S1 in the supplementary material). As expected, these embryos had disrupted craniofacial development (see Fig. S3 in the supplementary material). Furthermore, when we overexpressed Spry2 in neural crest, 100% of mutants had abnormal craniofacial structures but normal OFT morphology (see Table S1 and Fig. S3 in the supplementary material). This indicates that OFT septation is independent of FGF signaling directly to CNC. PDGF signaling, which is known to occur in neural crest (Richarte et al.,2007), is therefore not a major target of inhibition by sprouty 2 in these cells.

Extensive published evidence demonstrates that FGFs stimulate migration and invasiveness of endothelial cells (Auguste et al., 2003). Since the proximal OFT cushions of Fgf8;Isl1Cre mutants have few endothelium-derived mesenchymal cells,we examined the ability of endothelial cells from Fgf8;Isl1Cremutants to migrate from an explanted OFT, undergo EMT, and invade a collagen gel. Remarkably, 100% of mutant OFT explants exhibited a dramatic failure of EMT (Fig. 3B, n=11). In control explants, only an occasional cell appeared rounded and on the gel surface (Fig. 3A), whereas in the mutants, the few cells that migrated away from the explant were rounded up on the surface or fragmented into debris(Fig. 3B). For quantitation of these findings, see Fig. S4 in the supplementary material.

We then tested whether loss of FGF signaling directly to endothelial cells causes OFT defects using Tie2Cre(Kisanuki et al., 2001) to ablate Fgfr1 and Fgfr2. Tie2Cre conditional mutants survived and had normal OFTs (see Table S1 in the supplementary material; data not shown). This suggests that the endothelial EMT defect seen in Fgf8;Isl1Cre mutants in vivo (Fig. 1), and in explants (Fig. 3), is endothelial cell-non-autonomous. To test this, we co-cultured Fgf8;Isl1Cre mutant explants adjacent to those from controls. To identify the origin of invading cells in co-culture, we used the conditional Rosa26^lacZ line and Isl1Cre to label endothelial cells from control explants, or carried the Rosa26^lacZ allele in the females used to generate conditional mutant explants [the Isl1Cre expression domain includes OFT endothelium (Park et al.,2006) (data not shown)]. We found that control explants rescued the ability of Fgf8;Isl1Cre mutant endothelial cells to undergo EMT(Fig. 3D,E), with no adverse effect of lacZ expression in controls(Fig. 3C). In addition to confirming that the EMT defect in Fgf8;Isl1Cre mutants is not due to primary endothelial dysfunction, this finding suggests that the essential defect in the mutant OFTs resides within the myocardium itself.

Based on the co-culture results, we predicted that loss of FGF signaling to mesodermal precursors of the OFT would phenocopy the defects of Fgf8conditional mutants. We ablated Fgfr1 and Fgfr2 using the mesoderm-specific driver Mesp1Cre. This driver is active in the anterior primitive streak and ablates gene function in all myocardial and endocardial precursors (Park et al.,2006; Saga et al.,1999). Fourteen percent of Fgfr1^c/c;Fgfr2^+/+;Mesp1Cre mutants had alignment defects (TGA and DORV, Table 1). The frequency of alignment defects increased to 40% with decreasing Fgfr1/2 gene dosage in Fgfr1^c/c;Fgfr2^c/+;Mesp1Cre mutants(Fig. 4C, Table 1). Normally, persistence of myocardium in the right ventricular subvalvar outflow region (called the conus, derived from the SHF) and regression of this myocardium on the left side causes the pulmonary valve to ultimately reside above (distal to the ventricle) the aortic valve (Fig. 4A′). Hypoplasia of the right conus in Fgfr1/2mutants was apparent externally as a flattening of this region(Fig. 4C), and resulted in a side-by-side valve position (Fig. 4C′). Linkage of Mesp1 and Fgfr2 on chromosome 7, with a predicted recombination frequency of 23%, prevented analysis of large numbers of Fgfr1^c/+;Fgfr2^c/c;Mesp1Cre and Fgfr1^c/c;Fgfr2^c/c; Mesp1Cre mutants. However,there were no OFT defects in the Fgfr1^c/+;Fgfr2^c/c;Mesp1Cre mutants we obtained(Table 1). By contrast, four out of five of Fgfr1^c/c;Fgfr2^c/c;Mesp1Cremutants had PTA type III, a severe grade of PTA in which the OFT is unseptated along its entire proximodistal extent (Fig. 4D,D′). The truncal OFT valve was also abnormally aligned. Since the incidence and severity of OFT defects increase with decreasing Fgfr1/2 gene dosage in the mesoderm, the ability to align and septate the OFT depends on the amount of FGF signaling received by these cells. Our data also indicate that the dominant receptor transducing the required signals in this process is FGFR1, as FGFR2 has only partial compensatory activity.

Our observation of 100% type III PTA in Fgf8;Isl1Cre ligand mutants suggests that FGF8 signaling to mesoderm and endoderm might have additive influences on OFT remodeling. Indeed, ablation of Fgfr1 and Fgfr2 in the Isl1Cre domain [which encompasses both endoderm and SHF mesoderm at E8.0 (Park et al.,2006)] increased the frequency of type III PTA to 100% in Fgfr1^c/c;Fgfr2^c/c;Isl1Cre mutants(Table 1), as compared with 66%in mesoderm-only double mutants. Defects in Fgfr1^c/c;Fgfr2^c/+;Isl1Cre mutants were also more severe than those in their mesoderm-only counterparts because proximal OFT septation frequently failed, resulting in type I PTA(Fig. 4G, Table 1). In this form of PTA(which is common in human patients), there is a single truncal valve, but the distal OFT is septated (Fig. 4G,G′). Consistent with our hypothesis that FGFR1 is the principal receptor, Fgfr1^c/+;Fgfr2^c/c;Isl1Cremutants had milder defects than Fgfr1^c/c;Fgfr2^c/+;Isl1Cre mutants(Fig. 4F, Table 1).

These data suggest that FGF signaling to the pharyngeal endoderm plays a role in OFT remodeling. We used a novel Cre driver, Foxa2IRES mER-Cre-mER (Park et al.,2008), to ablate Fgf8 or Fgfr1 and Fgfr2 in pharyngeal endoderm by E8.0 and test whether independent effects of endodermal FGF signaling on OFT septation could be detected. Although loss of endodermal receptor function disrupts vascular development(E.J.P. and A.M.M., unpublished), endodermal ligand and receptor mutants survived to birth and had normal OFTs (see Table S1 in the supplementary material).

Antagonism of FGFR-mediated signaling in SHF mesoderm in Spry2-GOF;Mesp1Cre embryos caused phenotypes consistent with the receptor ablation results: 75% of these mutants had heart defects, most of which were arterial pole abnormalities (PTA and DORV, Fig. 5A-I, Table 1). Additionally, as in Fgf8;Isl1Cre mutants (Fig. 1), the OFTs of E9.5 Spry2-GOF;Mesp1Cre mutants were significantly shorter than in controls and were at an obtuse angle to the right ventricle (Fig. 5J-O and data not shown). Wnt11 transcripts in the OFT myocardium were reduced in E9.5 Spry2-GOF;Mesp1Cre mutants(Fig. 5J,K) and Bmp4expression was also notably downregulated in the OFT and SHF(Fig. 5L,M). Both Spry2-GOF;Mesp1Cre and FGFR mutants phenocopied the early defects in endothelial EMT and CNC invasion seen in Fgf8;Isl1Cre mutants(Fig. 5P,Q; see Fig. S5 in the supplementary material). Reduction in EMT was demonstrated in Spry2-GOF;Mesp1Cre mutant explants(Fig. 5R,S), similar to that shown for Fgf8;Isl1Cre mutants(Fig. 3A,B). Proliferation of ISL1-expressing cells in the SHF and OFT myocardium was significantly decreased in Spry2-GOF;Mesp1Cre mutants at multiple stages(Fig. 5T-V and data not shown);this is also seen in Fgf8 mutants(Park et al., 2006). Mesodermal Spry2-GOF phenotypes are similar to those of the FGFR mutants; since SPRY2 inhibits signaling downstream of other RTKs, this demonstrates the primacy of FGF signaling in SHF mesoderm for subsequent OFT morphogenesis. However, the effects on Wnt11 and Bmp4transcripts suggest that downstream modulation of other signaling pathways is likely to affect CNC and endothelial cell behavior in the OFT. Detection of these OFT markers, albeit at a lower level, demonstrates that OFT progenitors continue to be added from the SHF when FGF signaling is compromised.

In order to examine downstream effects of loss of FGF8 on myocardial signaling, we examined gene expression in isolated E9.5 OFTs from Fgf8;Isl1Cre mutants and controls using genome-wide microarray analysis. This timing allowed us to examine myocardial and endocardial gene activity at the onset of endocardial EMT and CNC invasion. In the FGF pathway(Fig. 6A), few genes were dysregulated greater than 2-fold. Expression of the known FGF8 target genes Pea3 and Erm was significantly decreased, as expected. Reductions seen in the expression of genes encoding FGF counter-regulatory factors [sprouty, Spred, Il17rd (Sef) and Dusp genes] were similar.

When FGF signaling to the SHF and endoderm was perturbed, there were striking changes in the level of transcripts of components of the BMP,TGFβ and semaphorin/plexin pathways, which are known to be essential to both endocardial EMT and CNC survival and invasion of the OFT cushions(Barnett and Desgrosellier,2003; High and Epstein,2007) (Fig. 6B,D,Eand data not shown). Quantitative RT-PCR confirmed these findings and revealed that the array frequently underestimated the magnitude of the expression changes (see Table S2 in the supplementary material). Bmp4 and Bmp2 transcripts were decreased, as was the expression of genes in the Bmp4 synexpression group(Karaulanov et al., 2004),including the transcriptional effectors Msx1 and Msx2, and target genes with roles in OFT endocardial EMT (Has2, Gata4, Tbx3, Twist1,Snai1) (Liu et al.,2004). Transcripts for some BMP antagonists (noggin, Bambi) and inhibitory Smads (Smad6, Smad7) were also decreased, and increased expression of the BMP antagonist gremlin 2 in the face of decreased Bmp4/2 would further decrease BMP signaling. Tgfb1 and Tgfbr1 were also downregulated, together with Tgfbrap1, a protein induced by TGFβ that transmits signal from TGFβ receptors to SMAD2 and 4(Wurthner et al., 2001).

BMP and TGFβ signaling regulate, and are modulated by, ECM composition. We found altered expression of genes encoding structural and signal-modulating ECM components. These included: Has2, a synthetic enzyme for hyaluronic acid, which is a major component of the cardiac jelly required for EMT (Camenisch et al.,2002); latent TGFβ-binding protein 1 (Ltbp1), a regulator of TGFβ bioavailability required for OFT septation(Todorovic et al., 2007); and the proteoglycans decorin (Dcn), biglycan (Bgn) and versican(Vcan; Cspg2), which influence the fibrous structure of collagen in the ECM and its physical interaction with cells, as well as the bioavailability and cellular responses to TGFβ and BMPs(Fig. 6B)(Macri et al., 2007).

Altered expression of these BMP/TGFβ pathway genes disrupted the activity of effector SMADs in the OFT cushions. The number of cells with detectable nuclear phosphorylated R-SMAD1/5/8 in the OFT endothelium and subendothelial mesenchyme of Fgf8;Isl1Cre and Spry2-GOF;Mesp1Cre mutants at E9.5 was dramatically decreased(Fig. 6G,I,I′). Our results clearly demonstrate that loss of FGF signaling negatively regulates signaling pathways that control EMT and CNC invasion of the OFT cushions.

In order to dissect how intercellular signaling regulates the complex cellular behaviors required for normal morphogenesis, it is crucial to identify not only the sources of the signals, but their cellular targets. In this way, we can discover immediate events and direct regulatory interactions in the target cells and distinguish them from secondary processes. In this study, we show that contrary to expectations, neural crest and endothelium are not the paracrine targets of the FGF signals that arise in SHF mesoderm and pharyngeal endoderm and are required for correct OFT formation, alignment and septation. Rather, we find that autocrine FGF signaling within the mesoderm of the SHF is absolutely required for septation of the OFT, and that signaling to the pharyngeal endoderm, although not independently required, has an additive function. Our data show that loss of FGF signaling within SHF mesoderm causes primary dysfunction of its derivatives in the OFT myocardium, evident by decreased production of cardiac jelly and dysfunction of pivotal signaling pathways from the myocardium to endothelium and CNC. Our results are consistent with those presented in a companion study(Zhang et al., 2008), in which ablation of the FGFR adaptor protein FRS2α (FRS2 - Mouse Genome Informatics) in the SHF and pharyngeal endoderm (but not neural crest or endothelium) causes arterial pole defects.

In our studies, FGFR gene ablation and Spry2 gain-of-function obtained with Mesp1Cre occur not only in mesodermal precursors of the OFT in the SHF, but also in their derivatives. However, Fgf8 ablation using Cre drivers with different temporal onset in the mesoderm (as it emerges from the primitive streak, at the crescent stage, when the SHF cells reside in the splanchnic mesoderm, or when SHF derivatives have already accrued to the heart) indicate that OFT defects only occur if Fgf8 function is ablated at or before the crescent stage(Brown et al., 2004; Ilagan et al., 2006; Park et al., 2006). This temporal window indicates that the required FGF/FGFR signaling event occurs in the mesodermal precursors of the OFT in the SHF. Similar temporal requirements were identified for FRS2α-mediated signaling(Zhang et al., 2008). Furthermore, decreased cell proliferation, increased apoptosis and alterations in the expression of Isl1 and known FGF8 target genes are already marked in the SHF mesoderm of Fgf8;Mesp1Cre and Fgf8;Isl1Cremutants by the 5- to 8-somite stage (Park et al., 2006); decreased proliferation is also striking in Spry2-GOF;Mesp1Cre embryos (Fig. 5), all of which point to primary effects on the progenitor cell population resulting in decreased OFT size and other defects.

An NKX2.5-BMP2/SMAD1 negative-feedback loop has been documented in the SHF in which diminution of BMP2/SMAD1 signaling increases progenitor cell proliferation and myocardial specification(Prall et al., 2007). Notably, Fgf8 function was preserved in this system. Our findings suggest that BMP4/FGF8 participate in a different regulatory pathway because compromising FGF signaling resulted in decreased proliferation in the SHF associated with decreased Bmp4 transcript levels in the SHF and reduced BMP signaling in the OFT. Indeed, differences between BMP2 and BMP4 function in the context of myocardial progenitor specification/proliferation, and in the OFT itself,have been documented (Armstrong and Bischoff, 2004; Klaus et al.,2007; Ma et al.,2005). In addition to effects on BMP signaling, which may affect the recruitment of differentiating myocardial cells(Waldo et al., 2001), we show that disrupted FGF signaling leads to reduced expression of Wnt11 in OFT myocardium (Fig. 5), which was also seen in our microarray analysis (not shown), consistent with previous observations on Fgf8;Isl1Cre mutants(Park et al., 2006). Perturbation of non-canonical WNT11 signaling in the myocardium affects OFT development and interferes with Tgfb2 transcription and ECM composition (Zhou et al.,2007).

The cushion defects in our FGF mutants focused our attention on the recruitment of neural crest. CNC cells do not arrive in the pharynx until∼E9.0. Thus, our finding that these cells are not the direct target of the early FGF signals is consistent with the temporal window identified using FGF ligand mutants in the SHF (Ilagan et al.,2006; Park et al.,2006). Although we did not specifically rule out a role for Fgfr3 or Fgfr4 in CNC during OFT remodeling, Fgfr3/4 mutants have no cardiovascular defects(Weinstein et al., 1998). Spry2-GOF expression in these cells would antagonize signal transduction downstream of all four FGFRs, and Spry2-GOF neural crest cells have normal OFTs. Factors that antagonize SPRY2 activity could also influence the phenotype of these mutants. However, in an accompanying report,Zhang et al. show that ablation of Frs2α in CNC does not disrupt OFT development (Zhang et al.,2008), which is consistent with our findings. Indeed, the severe OFT defects in Mesp1Cre and Isl1Cre receptor mutants, in which Fgfr1/2 function remains intact in CNC, indicate that disrupted FGF signaling to CNC is not required to generate these phenotypes.

Several lines of evidence indicate that secondary effects on CNC after loss of FGF signaling in the SHF play a role in the OFT phenotypes obtained in our mesodermal FGFR and Spry2 gain-of-function mutants. The finding of aortic arch malformations in these mutants (Figs 4 and 5), but not in the neural crest conditional mutants (see Fig. S3 in the supplementary material), is consistent with downstream effects on neural crest. Excessive neural crest apoptosis is observed in Fgf8 mutants (Ilagan et al., 2006; Macatee et al.,2003; Park et al.,2006), and Spry2-GOF;Mesp1Cre mutants have decreased neural crest Crabp1 expression. Aortic arch defects have also been shown to be the result of secondary neural crest dysfunction due to Tbx1 loss-of-function in the mesoderm of the SHF(Xu et al., 2004; Zhang, Z. et al., 2006). BMP and TGFβ signaling are essential for CNC invasion of the OFT cushions and for pharyngeal arch artery development(High and Epstein, 2007; Liu et al., 2004; Stottmann et al., 2004); the changes that we document in these pathways in the face of compromised FGF signaling will impact these processes. Modifications to ECM components, which also affect BMP/TGFβ signaling (Macri et al., 2007), further contribute to effects on the neural crest. In addition to the ECM and other signaling defects discussed above,transcription of Acvr1 (Alk2), semaphorin 3c, plexin A3 and neuropilin 2 is decreased in Fgf8;Isl1Cre mutants (not shown). Each of these proteins critically impacts CNC function(High and Epstein, 2007).

Neural crest cells modulate FGF8 signaling in the pharynx and influence not only the addition of myocardium to the OFT from the SHF(Hutson et al., 2006), but also the contractile and secretory function of the myocardium itself,including its ability to produce cardiac jelly(Stottmann et al., 2004; Waldo et al., 1999). Thus, in affected Fgf8, FGFR and Spry2 gain-of-function mutants,initial myocardial dysfunction and subsequent abnormal CNC behavior might interact in a cycle that progressively impairs OFT morphogenesis.

Since FGFR ablation in the endothelial precursors of the OFT endocardium does not perturb OFT remodeling, and EMT defects in Fgf8;Isl1Cremutants can be rescued by wild-type myocardium, direct signaling between FGF ligands produced by the SHF and the endocardium is not required for OFT septation. Furthermore, ablation of Frs2α in the endothelium does not disrupt OFT morphogenesis (Zhang et al., 2008). However, defects in the expression of ECM components and signal modulators downstream of the BMP/TGFβ pathways provide insight into the molecular basis of the secondary endothelial dysfunction we observe (Barnett and Desgrosellier, 2003; Liu et al., 2004). Effects on these signaling pathways resulting from loss of FGF8 compromise the expression of numerous target genes with roles in endocardial EMT (Has2, Gata4, Tbx3, Twist1, Snai1)(Armstrong and Bischoff, 2004; Liu et al., 2004). TGFβi(TGFβ induced) stimulates endothelial migration by altering the structure of VE-cadherin intercellular junctions and integrin activity(Ma et al., 2008), and its downregulation might contribute to defective EMT in the mutant OFT.

In contrast to the paradigm of paracrine signaling established in other tissues, our data show that in the SHF, the cellular source of the ligand(signal) is also the target. Such an autocrine pathway can be easily understood in terms of a feedback loop that maintains FGF production within a tight range (E.J.P. and A.M.M., unpublished), which is crucial for FGF8 function (Hutson et al.,2006). Secondary effects on other signaling pathways that we observe may also be integrated into this regulatory loop. Fgf8 and FGFR mutant analyses establish that the autocrine pathway not only regulates survival and proliferation of SHF cells (a common response to FGFs), but also the secretory and signaling capacities of their derivatives in the OFT (this study) (Zhang et al., 2008; Ilagan et al., 2006; Park et al., 2006). The few transcriptional targets of the PEA3 family of FGF8 effector proteins thus far identified are ECM components, ECM-modifying enzymes and cell adhesion molecules (de Launoit et al.,2006), suggesting that an autocrine pathway might provide a means of regulating the ECM and microenvironment to ensure uniform signal reception and response within a specialized cell population. Our findings are of biomedical importance, not only in the context of understanding the causes of congenital malformations of the OFT, but also because the crucial role we demonstrate for an autocrine FGF signaling pathway has broad implications for understanding fundamental properties of FGF signaling in different developmental and pathological contexts.

We thank C. Deng, J. Rossant, D. Ornitz, G. Martin, Y. Saga, J. Epstein and S. Evans for mouse lines; Brett Milash and the University of Utah Microarray Core; K. Thomas, M. Stevens and S. Vincent for critically reading the manuscript; and M. Stevens, D. Broka, E. Pecnard and C. Bodin for technical assistance. Y.W. and M.B. thank members of their lab, N. Duval, C. St Clement,R. Kelly and S. Zaffran for helpful discussions. S.M.-T. was supported by the Naito Foundation; Y.W. and M.B.'s lab by the E.U. Integrated Project `Heart Repair' (LHSM-CT2005-018630), the Pasteur Institute and the CNRS; E.M. and J.K.'s lab by NIH HDO4803; and E.J.P. and A.M.M.'s lab by NIH/NICHD, AHA and the University of Utah Program in Human Molecular Biology and Genetics.