Wnt/β-catenin signaling enables developmental transitions during valvulogenesis

Heart valve defects are remarkably common, affecting 2% of the population (Pierpont et al., 2007; Go et al., 2013). Many valve defects do not become pathologic until late adulthood, contributing to the high prevalence of valve replacement surgeries (Brickner et al., 2000). However, as with severe valve defects that require childhood intervention, the underlying abnormalities originate from disruption of embryonic valvulogenesis.

The mammalian heart has four sets of valves. The two semilunar valves (SLVs) – the aortic valve (AV) and pulmonic valve (PV) – have three symmetric hemi cup-like cusps. The cusps meet to form tight seals that only open under systolic pressure, allowing unidirectional blood flow from the ventricles into the systemic and pulmonic circulations, respectively. The mitral valve (MV) and tricuspid valve (TV) form the two atrioventricular canal (AVC) valves that support blood flow between the atria and ventricles. These valves have broad flap-like leaflets that attach to the ventricular walls by chordae tendineae (CT).

The SLVs develop from two sets of endocardial cushions, the proximal and distal outflow tract (pOFT and dOFT) cushions, that form in the unseptated arterial pole of the looping heart. Concurrently, AVC valve development initiates from two endocardial cushions at the junction of the common atrium and ventricle. Mesenchymal cells populate the endocardial cushions soon after they form. AVC cushion mesenchyme is derived by the endocardial-to-mesenchymal transformation (EMT) of overlying endocardial cells signaled to delaminate and invade the cardiac jelly of the cushions (Markwald et al., 1977). pOFT cushion mesenchyme also is largely EMT derived, whereas the dOFT cushions are primarily populated by cardiac neural crest cells (NCCs) (Kirby et al., 1983; Waldo et al., 1998; Jiang et al., 2000). EMT initiates around embryonic day (E) 9.5 and E10.0 in the AVC and pOFT, respectively (Camenisch et al., 2002).

EMT is a multi-step process. First, the cushion fields are specified by Notch and BMP signaling to generate specialized myocardium that produces EMT-inducing ligands and a correspondingly responsive endocardium (Timmerman et al., 2004; Luna-Zurita et al., 2010; Sugi et al., 2004; Ma et al., 2005; Wang et al., 2005; McCulley et al., 2008). TGF-β and other factors then induce changes in endocardial gene expression programs that repress epithelial-state determinants and upregulate mesenchymal factors that promote ECM invasion and cell migration (reviewed by MacGrogan et al., 2014). The canonical Wnt signaling pathway, involving transcriptional changes driven by stabilized nuclear β-catenin and TCF/Lef transcription factors, might also directly promote EMT (Liebner et al., 2004). However, it is unclear whether the loss of AVC mesenchyme in endothelial β-catenin (Ctnnb1)-null mice results from disrupted Wnt signaling or the distinct adherens junction role of β-catenin (Gottardi and Gumbiner, 2004; reviewed by Heuberger and Birchmeier, 2010). Earlier Wnt/β-catenin signaling roles in cardiac progenitor cells (Ai et al., 2007; Klaus et al., 2007) raises the additional possibility that Wnt/β-catenin signaling helps establish EMT-competent cushion fields.

Following EMT, cushion mesenchyme undergoes controlled proliferation in response to mitogenic signals. Wnt reporter transgenic mice suggest that canonical Wnt signaling is active in AVC and OFT cushion mesenchyme during this growth phase (Gitler et al., 2003; Alfieri et al., 2010; Klaus et al., 2012; Gillers et al., 2015; Cai et al., 2013; Cambier et al., 2014). Known mitogenic roles of Wnt/β-catenin signaling in other developmental contexts (reviewed by Niehrs and Acebron, 2012) further support a role for Wnt in cushion expansion. However, no Wnt loss-of-function studies in mice have defined a proliferative or other role of Wnt signaling in cushion mesenchyme.

As development proceeds, the valve primordia formed by the cushions elongate into thin cusps (SLVs) or leaflets (AVC valves) (reviewed by Hinton and Yutzey, 2011). The cusps/leaflets become stratified into layers termed the ventricularis/atrialis, spongiosa and fibrosa. Each layer acquires unique biomechanical properties conferred by the expression of distinct ECM proteins. Elastin confers flexibility to the atrialis and ventricularis layers of the AVC valves and SLVs, respectively (Vesely, 1997). The interconnecting spongiosa layer is proteoglycan enriched, notably with versican (Vcan) protein (Henderson and Copp, 1998; Mjaatvedt et al., 1998). Organized collagen fibrils provide rigidity to the fibrosa layer (Vesely, 1997; Lincoln et al., 2004). Additionally, the basal cusps of the semilunar valves are rich in tenascin-C (Tnc) (Akerberg et al., 2015). Similarly, the AVC valves have high Tnc expression where they attach to the ventricular wall, including in the chordae tendineae (CT) (Lincoln et al., 2004). Disrupted ECM patterning is one of the primary hallmarks of valve disease. While these later stages of valve formation are poorly understood, expression studies suggest canonical Wnt signaling functions during the valve patterning phase (Alfieri et al., 2010).

Models proposing that canonical Wnt signaling promotes cushion EMT, EC expansion and valve patterning have not been definitively tested using loss-of-function approaches. Although several Wnt transcripts, notably Wnt4 and Wnt9b, are specifically expressed in valves (Wang et al., 2013; Alfieri et al., 2010), there are no genetic studies directly implicating Wnt ligands in valve development, probably because of redundancy between the 19 Wnt family members. Wnt proteins can also signal through ‘non-canonical’ pathways – roles that cannot be distinguished by conventional genetics. Therefore, conditional deletion of floxed Ctnnb1 is the predominant approach to disrupt Wnt signaling in mice. However, this method also abolishes the aforementioned contribution of β-catenin to adherens junctions that maintain epithelia. Furthermore, Cre-lox conditional gene deletions do not provide sufficient temporal control over pathway activity to study each of the proposed dynamic roles of Wnt during valvulogenesis.

We developed an inducible loss-of-function approach using transgenic mice that provides spatiotemporal control over Wnt/β-catenin signaling during valve development. We use cell type-specific inducible expression of Dkk1, a secreted protein that binds the Wnt Lrp5/6 co-receptors to potently and specifically attenuate canonical Wnt signaling (Bafico et al., 2001; Mao et al., 2001; Semënov et al., 2001). Using this approach and refined expression studies of Wnt-responsive target genes, we show that Wnt has multiple roles throughout valve development. First, Wnt/β-catenin instructs OFT myocardium to produce EMT-inducing ligands while, surprisingly, not being directly required for EMT in either the pOFT or AVC. Second, Wnt promotes cushion expansion by supporting the responsiveness of mesenchyme to FGF signaling. This role as a mitogenic competence factor is progressively suppressed by feedback inhibition through accumulated expression of the negative Wnt regulator, Axin2. Third, Wnt/β-catenin signaling enables MV cushion mesenchyme to respond to subsequent ECM patterning cues that promote the spongiosa layer and restrain the CT-leaflet boundary. We propose that Wnt/β-catenin signaling enables transitions between the cushion establishment, expansion and patterning stages of valve development rather than directly inducing discrete developmental processes.

β-catenin, the transcriptional effector of canonical Wnt signaling, is required in the endothelial lineage for endocardial cushion EMT (Liebner et al., 2004). However, Ctnnb1 loss-of-function studies do not distinguish between its Wnt signaling and cell adhesion roles. To examine canonical Wnt signaling roles in developing mice, we established an inducible Tet-On transgenic system that provides spatiotemporal Wnt inhibition through Dkk1 expression (Fig. S1A). Dkk1 is a secreted, potent and specific inhibitor of canonical Wnt signaling that binds to the Lrp5/6 obligate co-receptors of Wnt ligands (Bafico et al., 2001; Mao et al., 2001; Semënov et al., 2001). Actin:rtTA;TRE:Dkk1 embryos exposed to doxycycline (Dox) beginning at E8.5 and harvested at E10.5 had stunted limb buds and a reduced forebrain (Fig. S1B,C). These phenotypes recapitulate those in mice lacking the Wnt co-receptor Lrp6 (Song et al., 2009b). Further validating our approach, expression of Axin2, which is a well-established canonical Wnt target gene (Jho et al., 2002), was reduced in embryos with ubiquitous Dkk1 expression (Fig. S1B,C, inset). Likewise, Actin:rtTA;TRE:Dkk1 embryos had reduced levels of Lef1, which is transcriptionally upregulated by the canonical Wnt pathway (Filali et al., 2002), and nuclear β-catenin in limb bud tissue (Fig. S1D-G).

E10.5 embryos ubiquitously expressing Dkk1 from E8.5 lacked mesenchymal cells within the pOFT cushion, consistent with the proposed role of Wnt in cushion EMT (Fig. S2A,B). However, AVC EMT proceeded normally (Fig. S2C,D). Therefore, the AVC EMT deficiency in endothelial lineage Ctnnb1-deficient mice (Liebner et al., 2004) reflect adherens junctions roles of β-catenin in maintaining the epithelial integrity of the endocardium.

The pOFT EMT deficiency upon ubiquitous Dkk1 expression could be secondary to widespread developmental defects upon global Wnt inhibition. Therefore, we used a variant Tet-on system to specifically express rtTA in the endothelial lineage (Belteki et al., 2005). In Tie2:Cre;R26R:rtTA;TRE:Dkk1 embryos, endothelial-expressed Cre recombinase excises the loxP-Stop-loxP cassette of R26R:rtTA to drive rtTA uniquely in the endothelial/endocardial lineage (Fig. 1A). We validated the responsiveness and specificity of this system using a surrogate TRE:H2BGFP transgene (Tumbar et al., 2004; Fig. S3A). Dox-dependent nuclear GFP expression in endothelial, endocardial and EMT-derived mesenchymal cells of Tie2:Cre;R26R:rtTA;TRE:H2BGFP embryos required all three transgenes and was strong 6 h after injection (Fig. S3B-I).

E10.5 Tie2:Cre;R26R:rtTA;TRE:Dkk1 embryos exposed to Dox from E8.5 had quantitatively reduced pOFT mesenchyme and normal dOFT and AVC mesenchyme (Fig. 1B-F). Proliferation of AVC and residual pOFT mesenchymal cells, as assessed by BrdU incorporation, was unchanged in Dkk1-expressing embryos (Fig. 1F). Therefore, Wnt/β-catenin signaling is required to generate pOFT mesenchyme but is not required for initial cell proliferation following EMT in either set of cushions. As AVC EMT precedes pOFT EMT (Camenisch et al., 2002), we tested whether earlier induction of Dkk1 would uncover an AVC EMT defect upon Wnt inhibition. However, Dkk1-expressing embryos at E8.0-E10.5 still only showed reduced pOFT mesenchyme (Fig. S4A-D).

Although dOFT and pOFT mesenchyme are primarily neural crest cell (NCC) and EMT-derived, respectively, there is some co-mingling of the two lineages (reviewed by Snarr et al., 2008). Therefore, to confirm that induction of Dkk1 reduced EMT- and not NCC-derived pOFT mesenchyme, we introduced the TRE:H2BGFP transgene to simultaneously inhibit Wnt and trace endocardial-derived mesenchyme. As expected, Tie2:Cre;R26R:rtTA;TRE:H2BGFP;TRE:Dkk1 embryos expressing Dkk1 from E8.5 to E10.5 had significantly fewer GFP⁺ cells in the pOFT (Fig. S5A,B,E). Dkk1 induction had no effect on the number of GFP⁺ cells in the AVC and non-GFP⁺ cells (derived from NCC) in the dOFT (Fig. S5C-E).

To evaluate the direct requirement of canonical Wnt signaling in pOFT EMT, we performed EMT explant assays using Tie2:Cre;R26R:rtTA;TRE:H2BGFP;TRE:Dkk1 embryos. To mimic the in vivo experiments, we initiated Dox treatment in utero at E8.5 and explanted cushions at E9.5 onto Dox-infused collagen gels. Dkk1-expressing OFT but not AVC cushion explants exhibited a substantial decrease in the number of GFP⁺ mesenchymal cells (Fig. 1G-K). While canonical Wnt signaling therefore has a unique role in promoting pOFT EMT, the secreted nature of Dkk1 does not resolve whether Wnt signaling is required in the endocardium or in adjacent myocardium.

To further explore the role of canonical Wnt signaling in pOFT EMT, we revisited expression of transgenic and endogenous reporters of Wnt activity in the endocardial cushions. Previous surveys of Wnt/β-catenin activity in heart development have primarily used the TOPGAL and BATGAL transgenic reporters (Alfieri et al., 2010; Cai et al., 2013; Cambier et al., 2014; Gitler et al., 2003; Maretto et al., 2003) that are non-specific in some contexts (Al Alam et al., 2011; Barolo, 2006). Therefore, we examined patterns­ of Wnt signaling activity using a lacZ knock-in allele of Axin2, a well-established universal Wnt target gene (Jho et al., 2002; Lustig et al., 2002). Corroborating previous reports (Gillers et al., 2015; Jain et al., 2015; Klaus et al., 2012), sections of X-gal-stained Axin2^lacZ/+ E9.5 embryos showed faint Axin2 expression in pOFT but not dOFT myocardium, the ventricular side of AVC myocardium and AVC mesenchyme (Fig. 2A,B). At E10.5, the highest Axin2 expression persisted in pOFT and AVC myocardium, with lower levels in pOFT and AVC mesenchyme (Fig. 2C,D). We observed little to no Axin2 expression in endocardial cells of either cushion or in dOFT myocardium at either stage, arguing against a direct role for canonical Wnt signaling in promoting EMT.

Expression of the transcription factor Lef1 can mark both Wnt-responsive cells and, as a target gene, cells actively transmitting Wnt signals (Filali et al., 2002). We observed strong Lef1 expression in E9.5 pOFT and AVC myocardium and in AVC mesenchyme (Fig. 2E,F). By E10.5, Lef1 levels were decreased in both the pOFT and the AVC myocardium whereas high levels of Lef1 persisted in AVC and the now evident pOFT mesenchyme (Fig. 2G,H). These patterns largely recapitulate a previous report (Cai et al., 2013). Only sporadic endocardial cells showed low levels of Lef1 in both pOFT and AVC at E9.5 and E10.5, arguing, as with the Axin2^lacZ reporter, that canonical Wnt signaling does not directly induce EMT.

Consistent with active Wnt signal transduction, E9.5 pOFT myocardial cells expressed nuclear β-catenin (Fig. 2I), which was decreased in pOFT myocardium of E8.5-E9.5 Dox-treated Tie2:Cre;R26R:rtTA;TRE:Dkk1 embryos (Fig. 2J). Therefore, endocardial-induced secreted Dkk1 is capable of attenuating canonical Wnt activity throughout the cushion field. In contrast to its myocardial expression, β-catenin was predominantly membrane localized and non-nuclear in pOFT endocardium (Fig. 2I). Axin2, Lef1 and β-catenin expression all indicate that canonical Wnt signaling is mainly active in pOFT cushion myocardium during EMT. Therefore, the pOFT EMT defect upon Dkk1 induction probably reflects a cushion myocardial role for Wnt signaling in promoting neighboring endocardial cells to undergo EMT.

To test whether myocardial Wnt signaling promotes expression of a secreted EMT-inducing ligand, we performed EMT explant assays with E9.5 wild-type OFT tissue placed adjacent to a normally EMT-deficient Dkk1-expressing OFT. The Dkk1-expressing explant included the TRE:H2BGFP transgene to allow exclusive scoring of its EMT-derived cells otherwise intermingled with NCC-derived and wild-type explant cells. Wild-type OFTs robustly rescued EMT of immediately adjacent Tie2:Cre;R26R:rtTA;TRE:H2BGFP;TRE:Dkk1 Dox-exposed explants (Fig. 3A-D). Therefore, Dkk1-expressing endocardium remains competent to undergo EMT and Wnt/β-catenin signaling acts in neighboring pOFT myocardium to produce an EMT inducer(s).

We set out to determine when myocardial Wnt signaling is required for pOFT EMT by introducing Dox at different developmental ages. Initiating Dkk1 expression at E9.5 using Tie2:Cre;R26R:rtTA;TRE:Dkk1 embryos failed to reproduce the pOFT EMT seen with earlier inductions (Fig. S6A-C). Furthermore, Tie2:Cre;R26R:rtTA;TRE:H2BGFP;TRE:Dkk1 OFTs explanted on Dox-containing collagen gels without a priori in utero exposure underwent normal EMT (Fig. S6D-F). The lack of an EMT defect in these experiments is unlikely to represent delayed Dkk1 production, because endocardial lineage-traced wild-type OFTs explanted on collagen gel containing 100 nM Wnt-C59, a potent Porcupine inhibitor (Proffitt et al., 2013) that specifically prevents Wnt secretion, also completed EMT normally (Fig. S6G-I). Conversely, the EMT defect persisted when Tie2:Cre;R26R:rtTA;TRE:H2BGFP;TRE:Dkk OFTs exposed to Dox in utero at E8.5 were explanted onto Dox-free collagen gels (Fig. S6J-L). These timing experiments indicate that Wnt signaling acts between E8.5 and E9.0 to enable E9.5/E10.0 pOFT myocardium to produce an EMT-inducing substance (Fig. 3E). The early function of Wnt suggested an interface with either the Notch/BMP myocardial/endocardial crosstalk that render the cushion fields competent to undergo EMT. However, pOFT endocardial cells expressing Dkk1 from E8.5 to E9.5 retained Notch and BMP activity as monitored by anti-NICD and anti-pSmad1/5/8 staining, respectively (Fig. S7A-D). Further, transcript levels of Bmp2, Bmp4 and Tgfb2, another myocardial-expressed EMT-inducer (Potts and Runyan, 1989; Camenisch et al., 2002), were normal in Dkk1-induced hearts (Fig. S7E).

The upregulation of Axin2 and Lef1 in E10.5 OFT and AVC cushion mesenchyme (Fig. 2C,D,G,H) and previous Wnt reporter studies (Alfieri et al., 2010; Cai et al., 2013; Cambier et al., 2014; Gillers et al., 2015) indicate that Wnt signaling persists as cushions grow and elongate into mature valves. To explore Wnt signaling dynamics during the cushion/valve expansion phase, we assayed Wnt activity by Lef1 immunostaining. At E11.5, cushion myocardial Lef1 had largely dissipated (Fig. 4A-C). As previously described (Cai et al., 2013), occasional Lef1⁺ cushion endocardial cells were found in both the AVC and OFT (Fig. 4A-C). However, Lef1 was most robustly expressed in cushion mesenchyme of pOFT, dOFT and AVC cushions. By E13.5, Lef1 was primarily mesenchymal with enriched expression at the distal aspects of the forming aortic and pulmonic valve cusps and mitral valve leaflets (Fig. 4D-F).

To test Wnt/β-catenin signaling requirements during cushion/valve expansion stages, we induced global Dkk1 expression from E10.5 to E13.5 using Actin:rtTA;TRE:Dkk1 mice. These embryos developed small, abnormally shaped limbs and bilateral cleft lips (Fig. S8A-C), reminiscent of other canonical Wnt loss-of-function studies (Galceran et al., 1999; Song et al., 2009a,b). Lef1 protein was largely depleted in limb sections (Fig. S8D,E), underlining the efficacy of Dkk1 induction. Hearts from Actin:rtTA;TRE-Dkk1 embryos (Dox E10.5-E13.5) had ventricular septal defects, occasional double outlet right ventricle (DORV) and quantitatively smaller and blunted mitral valves (Fig. S8F-K). To determine whether these cushion/valve defects were a direct repercussion of Wnt attenuation in the heart, we used the Nfatc1^Cre line (Wu et al., 2012) with R26R:Stop-rtTA to drive endocardial lineage-specific Dkk1 expression. The TRE:H2BGFP reporter showed this approach produced Dox-dependent expression in nearly all endocardial cells, most AVC cushion mesenchymal cells and a smaller subset of OFT mesenchyme (Fig. S9). Unlike Tie2:Cre, Nfatc1^Cre is not active in endothelial cells outside the heart, reducing concerns of secondary effects from Wnt attenuation throughout the vascular system.

Nfatc1^Cre;R26R:rtTA;TRE:Dkk1 embryos exposed to Dox from E10.5 to E13.5 had quantitatively smaller and hypocellular MV leaflets (Fig. 4G-I). These endocardial lineage Dkk1-expressing embryos had 23% fewer BrdU-incorporating mesenchymal cells (Fig. 4I, P<0.04). Apoptosis, as assayed by cleaved caspase-3 (CC3) immunostaining, was unchanged (Fig. S10A-C). Therefore, canonical Wnt signaling supports the highly proliferative state of AVC cushion mesenchymal cells following EMT. Although Dox-exposed E13.5 Nfatc1^Cre;R26R:rtTA;TRE:Dkk1 embryos did not have a similar defect in their SLVs (Fig. S10D-F), this negative result is likely to reflect limited Nfatc1^Cre-driven recombination in SLV mesenchyme.

We used the inducible Dkk1-expressing system to determine when AVC cushion mesenchyme expansion requires Wnt/β-catenin signaling (results summarized in Fig. 4J). Nfatc1^Cre;R26R:rtTA;TRE:Dkk1 embryos exposed to Dox at E10.5-E11.5, E11.5-E12.5, E12.5-E13.5 or E12.5-E15.5 did not show gross morphological defects. However, AVC mesenchyme BrdU incorporation was decreased significantly in E10.5-E11.5 and E11.5-E12.5 Dkk1-expressing embryos (Fig. 4K). Therefore, canonical Wnt signaling is uniquely required for AVC mesenchyme proliferation between E10.5 and E12.5. When Wnt/β-catenin signaling was inhibited during this ‘execution period’, a reduced proliferative rate persisted to later stages. As such, E10.5-E13.5 Dkk1-expressing embryos gradually accumulated a deficit of AVC cushion mesenchymal cells, explaining why hypoplastic valves were only evident with longer Dox exposure. In agreement, E10.5-E15.5 Dox-exposed Nfatc1^Cre;R26R:rtTA;TRE:Dkk1 embryos also had smaller and blunted mitral valves (Fig. S10G-I). The partial proliferation block and delayed phenotype manifestation upon Wnt/β-catenin inhibition suggest a supporting rather than switch-like role for Wnt signaling in AVC cushion mesenchyme proliferation.

Receptor tyrosine kinases (RTKs) acting through Ras/MAPK signaling stimulate EC mesenchyme proliferation (Krenz et al., 2005, 2008). By staining for phosphorylated extracellular signal related kinase 1/2 (pErk1/2), we observed enriched Ras/MAPK activity in distal mesenchyme of E11.5 AVC cushions (Fig. 5A,B), coincident with highest Axin2 and Lef1 expression. E10.5 to E11.5 Dox-exposed Tie2:Cre;R26R:rtTA;TRE:Dkk1 embryos had markedly decreased pErk1/2 in AVC mesenchyme (Fig. 5C,D), suggesting that Wnt/β-catenin promotes AVC mesenchyme proliferation by enabling mitogenic RTK signaling.

Fibroblast growth factor (FGF) RTK signaling is implicated in cushion development (Park et al., 2006; Sugi et al., 2003; Zhang et al., 2008, 2010). Moreover, Wnt and FGFs coordinate cell growth in other developmental contexts (Stulberg et al., 2012; ten Berge et al., 2008; Yin et al., 2008). Either recombinant Wnt3A or Fgf1/2 increased the fraction of EdU-incorporating GFP lineage-labeled cushion mesenchymal cells cultured from Tie2:Cre;Rosa:rtTA;TRE:H2BGFP embryos (P<0.0001; Fig. 5E,F). The FGF receptor inhibitor PD-173074 alone reduced EdU incorporation rates (P<0.0001; Fig. 5E). Cells treated with both Wnt3A and PD-173074 showed a significant decrease in proliferation compared with cells treated with Wnt3A alone (P<0.003; Fig. 5E). Therefore, canonical Wnt signaling directs the competency of EC mesenchyme to robustly respond to growth-promoting FGF signaling.

The absence of proliferation defects when Wnt is inhibited after E12.5 suggests that Wnt activity is actively suppressed as cushions transition from the growth to elongation phase of valve development. As Axin2 is both a target gene and negative regulator of canonical Wnt signaling, we hypothesized that a negative feedback loop restrains cushion mesenchymal Wnt activity at later stages of valve development. We observed Axin2 expression, monitored using the Axin2^lacZ reporter allele, in E11.5 pOFT and AVC cushion myocardium and mesenchyme (Fig. 6A,B). At E13.5, Axin2 expression was largely restricted to cushion mesenchyme of the forming semilunar and AVC valves, represented by the pulmonic and mitral valves, respectively (Fig. 6C,D). By E18.5, Axin2 was strongly expressed throughout the mitral valves, as previously observed (Gillers et al., 2015) (Fig. 6E). This progressively stronger and broader Axin2 expression is consistent with negative feedback inhibition of Wnt signaling as valves exit the expansion phase of their development.

We used the Axin2^lacZ allele, which disrupts Axin2 transcription (Lustig et al., 2002), to test whether Axin2 functions as a negative regulator of Wnt signaling during later stages of embryonic valve development. Whereas Axin2^lacZ/lacZ embryos were present at the expected Mendelian ratio up to E16.5, homozygous null neonates were significantly underrepresented (9.4% observed versus 25% expected, n=53, P<0.009), and displayed characteristic craniofacial defects (Yu et al., 2005). E13.5 PV cusps of Axin2^lacZ/lacZ embryos had normal sectional areas and length-to-width ratios. However, PV cusp mesenchyme was mildly hyperplastic (Fig. 6F-H). By E16.5, PV cusp area, cell number and proliferation were all significantly increased in Axin2^lacZ/lacZ embryos (Fig. 6L, Fig. S11A,B). MV leaflets of Axin2^lacZ/lacZ embryos were significantly larger at E13.5 and contained more proliferative mesenchyme (Fig. 6I-K). While E16.5 Axin2^lacZ/lacZ embryos had unchanged MV leaflet sectional area and length-to-width ratios, cell counts and proliferation remained increased, suggesting a MV volume increase not evident by sectional analysis (Fig. 6M, Fig. S11C,D). 25% of E16.5 Axin2^lacZ/lacZ embryos (2/8 embryos) also had a membranous ventricular septal defect (VSD), consistent with underlying cushion abnormalities (Fig. S11E,F). Axin2 is therefore both a reporter of Wnt activity and a negative feedback regulator that helps restrict Wnt-promoted cushion growth to the expansion phase of valvulogenesis.

After expanding, cushions elongate into thin valve leaflets/cusps that become stratified into three distinct ECM layers (reviewed by Hinton and Yutzey, 2011). Expression studies implicate canonical Wnt signaling in valve ECM patterning during late embryogenesis and early postnatal development (Alfieri et al., 2010). Although our Axin2 expression studies suggest canonical Wnt is largely inhibited during late valvulogenesis, localized Lef1 expression in E13.5 MV leaflets (Fig. 4F) indicates that a subset of mesenchyme remains Wnt-responsive to pattern the ECM of developing valves. Supporting this possibility, Lef1 expression at E15.5 and E18.5 was concentrated in interstitial and rare endocardial cells at the mid-length of the extending MV leaflets as well as the forming CT (Fig. 7A,B).

Nfatc1^Cre;R26R:rtTA;TRE:Dkk1 embryos, exposed to Dox at E15.5-E18.5 or E13.5-E18.5, did not display grossly or morphometrically misshapen mitral valves (Fig. S12A-C). However, Movat's pentachrome staining showed that Dkk1-expressing MV leaflets had a reduced mucopolysaccharide-rich spongiosa layer (Fig. S13A,B). To further characterize Wnt-dependent MV patterning, we stained Dkk1-expressing embryos (E13.5-E18.5) for specific ECM proteins. Postn levels were unchanged (Fig. S13C,D). By contrast, the spongiosa component Vcan was notably decreased (Fig. 7C,D), which was also apparent in whole heart western blots (Fig. 7E). Dkk1 expression also caused Tnc to expand throughout the MV leaflet in contrast to its normal concentration at the base of the MV leaflets and CT (Fig. 7F,G). Increased Tnc levels were corroborated by western blot analysis (Fig. 7H). The shorter Dkk1 induction from E15.5 to E18.5 did not produce ECM changes in the MV of Nfatc1^Cre;R26R:rtTA;TRE:Dkk1 embryos (Fig. S13E,F). Therefore, canonical Wnt regulates MV ECM patterning between E13.5 and E15.5, preceding the period when the MV becomes evidently stratified. Therefore, akin to its roles in pOFT EMT and cushion growth, Wnt/β-catenin behaves as a competence factor that enables subsequent valve ECM patterning cues.

Our systematic functional study of Wnt/β-catenin signaling during mouse valve development extends and revises existing models (Fig. S14). First, we show that Wnt/β-catenin signaling does not directly promote endocardial cushion EMT, as previously surmised based largely on endothelial/endocardial-specific deletion of Ctnnb1. Rather, canonical Wnt signaling in pOFT cushion myocardium supports the production of an EMT-inducing molecule(s). Second, Wnt/β-catenin enables FGF-driven proliferation of AVC cushion mesenchyme following EMT. Negative feedback by accumulated Wnt-induced Axin2 in cushion mesenchyme progressively suppresses cushion growth as valves transition to the patterning phase of their development. Third, Wnt/β-catenin signaling in a subpopulation of AVC mesenchyme promotes formation of the Vcan-rich spongiosa layer and restricts Tnc expression to valve-muscle junctions, including the CT. In each of the establishment, expansion and patterning stages of valve development, the enabling function of Wnt/β-catenin precedes the immediate instructive cue. We propose that canonical Wnt signaling coordinates developmental transitions between steps of valvulogenesis that allow the progressive formation of these complex and congenital disease-prone tissues.

Our timing, expression and co-culture experiments indicate that canonical Wnt signaling acts indirectly to promote pOFT EMT. We only observed pOFT EMT defects when initiating Wnt inhibition at least a half day prior to EMT both in vivo and in vitro. Furthermore, Wnt activity in the OFT field at the onset of EMT was most evident in cushion myocardium, with little to no activity in cushion endocardial cells. Most importantly, adjacently placed wild-type OFT tissue rescued the EMT defect in Dkk1-expressing OFTs using explant assays. Therefore, Wnt/β-catenin-inhibited endocardium is fully competent to undergo EMT and Wnt-inhibited cushions only lack a secreted EMT-inducing substance(s). A cushion myocardial site of action for Wnt signaling is consistent with Wnt4 being sufficient to induce myocardial BMP2 expression and thereby promote AVC EMT (Wang et al., 2013). However, opposing that study's conclusions, myocardial β-catenin transcriptional activity is not required for EMT (Gillers et al., 2015) and we did not observe a Dkk1-induced EMT defect in the AVC cushions.

How does canonical Wnt activity in OFT cushion myocardium support EMT of the neighboring endocardial cells? In the AVC, EMT competency is conferred by myocardial/endocardial crosstalk that activates both Notch and BMP signaling (Luna-Zurita et al., 2010; Timmerman et al., 2004; Wang et al., 2005). However, NICD and pSmad1/5/8 staining showed that both pathways were intact in Dkk1-expressing OFT cushions. Tgfb2 expression levels were also unaffected by Dkk1 induction, suggesting that EMT-promoting TGFβ signaling is intact (Brown et al., 1996; Ramsdell and Markwald, 1997; Sridurongrit et al., 2008). Therefore, the Wnt-dependent EMT inducer(s) could be a novel EMT-driving factor. The source of the Wnt ligands that activate canonical Wnt signaling in OFT myocardium is probably the neighboring endocardium, given that these cells robustly express both Wnt4 and Wnt9b (Alfieri et al., 2010; Cai et al., 2013; Wang et al., 2013). This underscores the complex crosstalk between myocardium and endocardium required for EMT (de la Pompa and Epstein, 2012).

Our results argue against the widely reported conclusion that endocardial Wnt/β-catenin signaling is required for AVC EMT – a model based on potentially specious interpretations of two previous reports. First, the observation that Wnt-inhibited zebrafish embryos fail to produce primordial AVC cushions (Hurlstone et al., 2003) does not actually demonstrate an EMT defect because zebrafish AVC valve leaflets initially form through endocardial invagination (Scherz et al., 2008). Second, disrupted AVC EMT in mice with endothelial/endocardial deletion of Ctnnb1 might reflect non-transcriptional roles of β-catenin in forming adherens junctions that maintain the endocardium as an epithelium. Consistent with this idea, specifically disrupting the Wnt transcriptional roles of β-catenin in cardiomyocytes produces milder heart defects than complete loss of myocardial β-catenin (Gillers et al., 2015). Furthermore, deficient cushion Wnt/β-catenin activity upon conditional deletion of Tbx20 does not cause an AVC EMT defect, but instead restrains cushion growth (Cai et al., 2013). More broadly, we suggest conditional genetic studies of Ctnnb1 should be interpreted cautiously before assigning phenotypes to disrupted canonical Wnt signaling.

We observed that Wnt/β-catenin activity within cushion fields transitioned from primarily myocardial prior to and during EMT to almost entirely mesenchymal by E11.5. By using Nfatc1^Cre to induce Dkk1 in endocardial cells and EMT-derived cells, we show that the first role of Wnt in AVC cushion mesenchyme is to promote cell proliferation. In agreement, endocardial Tbx20 regulates Wnt signaling to direct cushion growth (Cai et al., 2013). We probably did not observe an SLV growth defect upon endocardial lineage Dkk1 expression because the Nfatc1^Cre line we used drives limited recombination in SLV mesenchyme. Although Dkk1 induction reduced AVC cushion mesenchyme, the effect was incomplete and resulting morphological changes relied on extended periods of canonical Wnt inhibition. These findings are most consistent with Wnt having a supporting mitogenic role rather than acting as a primary growth switch.

We show that canonical Wnt signaling promotes the responsiveness of cushion mesenchymal cells to mitogenic fibroblast growth factor (FGF) signaling. First, Ras/MAPK signaling monitored by pErk1/2 staining was reduced in Wnt-inhibited AVC mesenchyme. Second, recombinant Wnt-induced proliferation of cultured EMT-derived cushion mesenchymal cells was blocked by a specific FGF receptor inhibitor. A synergistic relationship between Wnt and FGF signals similarly promotes proliferation in the developing limb and lung mesenchyme (Stulberg et al., 2012; ten Berge et al., 2008; Yin et al., 2008). Therefore, Wnt might use a common mechanism to promote proliferation in developmental contexts (reviewed by Klaus and Birchmeier, 2008; van Amerongen and Nusse, 2009) by increasing FGF production or augmenting FGF responsiveness. Of disease relevance, Noonan syndrome, which is caused by mutations that result in excessive Ras/MAPK activity, is characterized by hyperplastic cushions leading to pulmonic stenosis (Tartaglia et al., 2001). Therefore, our study implicates excessive developmental Wnt/β-catenin activity as a potential contributor to Noonan syndrome-like congenital valve defects characterized by valve hyperplasia.

By expressing Dkk1 at various stages of valve development, we determined Wnt/β-catenin acts between E10.5 and E12.5 to promote cushion expansion. In a seeming contradiction, the Wnt target gene Axin2 was broadly expressed in cushion mesenchyme after E12.5 and in valve interstitial cells through postnatal stages (Fang et al., 2014). Axin2 is a potent cell-intrinsic inhibitor of Wnt signaling by providing a scaffold for the cytosolic protein destruction complex involved in β-catenin catabolism. Axin2-null mice developed large valves and showed increased cushion mesenchyme proliferation as early as E13.5. We propose that Wnt/β-catenin initially transcriptionally activates Axin2 expression in a restricted pool of cushion mesenchymal cells. These cells maintain Axin2 levels independent of continuous Wnt/β-catenin input and gradually populate most of the valve interstitium. Accumulated Axin2 constrains growth-promoting canonical Wnt signaling, accounting for restricted Lef1 expression at later stages. This negative feedback establishes a self-timing mechanism whereby accumulated Wnt activity supports the transition from the cushion growth to patterning phases of valve development. A similar Wnt/Axin2 feedback repressor circuit moderates tissue growth during skull development (Yu et al., 2005), implying that this mechanism is a common logic component of organogenic networks. Our study also highlights that, like wholly synthetic Wnt/β-catenin reporter lines, the Axin2^lacZ allele provides an imperfect means to monitor sites of active canonical Wnt signaling.

Canonical Wnt activity in AVC cushion mesenchyme monitored by robust Lef1 staining becomes progressively concentrated in the medial aspect of the mitral valve leaflets, coinciding with the region of abundant proteoglycan-rich spongiosa. Correspondingly, we observed that Dkk1-expressing mitral valves developed a reduced and Vcan-deficient spongiosa layer. Vcan is a known Wnt target gene (Rahmani et al., 2005) and therefore its misexpression could represent a direct effect of Wnt inhibition. Tnc expression, which is normally restricted to the rigid leaflet-muscle attachment sites including the CT, expanded to replace the diminished spongiosa in Dkk1-expressing embryos. Therefore, we anticipate that Wnt/β-catenin-inhibited valves would become inflexible, potentially leading to valve stenosis and regurgitation.

We saw no appreciable ECM patterning defects when canonical Wnt signaling was inhibited after E15.5, indicating that Wnt promotes ECM patterning prior to the first evidence of stratification. Therefore, canonical Wnt establishes tissue responsiveness to subsequent ECM patterning cues, which could include biomechanical forces provided by blood flow (Hove et al., 2003; Butcher et al., 2007), rather than directly instructing stratification. Although conditional deletion of Ctnnb1 in a subset of valve interstitial cells contradictorily does not affect valve growth or patterning (Fang et al., 2014), the Postn:Cre line used in that study might not promote a sufficient loss of Ctnnb1 and/or at an early enough stage to disrupt Wnt's pre-patterning role.

Could a common mechanism underlie the seemingly disparate pOFT EMT, AVC cushion expansion and mitral valve ECM patterning roles of canonical Wnt signaling? Wnt signaling is emerging as a universal progenitor cell maintenance factor that opposes cell differentiation (reviewed by Holland et al., 2013). During valvulogenesis, Wnt/β-catenin could first act to impede OFT cushion myocardium from fully differentiating into mature muscle. As such, OFT myocardial cells would retain functional aspects of their prior status as second heart field progenitor cells, in which canonical Wnt signaling promotes expression of growth factors that could include EMT-inducing ligands (Ai et al., 2007; Klaus et al., 2007). In developing cushion/valve mesenchyme, active Wnt signaling could maintain a progressively restricted pool of undifferentiated mesenchymal cells that produce a growth-supportive Vcan-rich microenvironment and remain responsive to additional developmental cues like FGF. As myxomatous valves are characterized by increased levels of Vcan and other proteoglycans (Gupta et al., 2009), excessive Wnt activity could promote this common manifestation of valve disease. Postnatally, specialized Wnt-responsive valve interstitial cells might serve as a stem-cell reservoir for valve homeostasis and repair. This possibility is supported by adult-onset valve disease upon Ctnnb1 deletion in a subset of cushion mesenchyme (Fang et al., 2014).

We propose that Wnt/β-catenin does not induce discrete events of valve development, but instead serves as a developmental timer that integrates with directly instructive signals. The molecular mechanism might not be through archetypal gene expression changes of a narrow set of target genes. Rather, Wnt signaling roles might promote, for example, a cellular metabolic state (Esen et al., 2013), or asymmetric cell divisions (Habib et al., 2013) that generally curtail differentiation. Regardless, the consequences of canonical Wnt inhibition on valve development highlight how both increased and decreased Wnt activity could cause various manifestations of congenital valve disease.

Mouse procedures were approved and monitored by the University of Oregon's Institutional Animal Care and Use Committee. TRE:Dkk1 mice carry a transgene constructed by cloning a HA-tagged mouse Dkk1 cDNA into the pTRE-Tight plasmid along with a 5′ signal peptide sequence. A founder line was selected that produced robust Dox-inducible Dkk1 expression and expected embryonic Wnt loss-of-function phenotypes in combination with Actin:rtTA (from Dr Steven Artandi; Sarin et al., 2005). Nfatc1^Cre mice were shared by Dr Bin Zhou (Wu et al., 2011, 2012). Tie2:Cre (Kisanuki et al., 2001), Axin2^lacZ (Lustig et al., 2002), R26R:rtTA (Belteki et al., 2005) and TRE:H2BGFP (Tumbar et al., 2004) mice were acquired from the Jackson Laboratory. Embryos were collected from timed matings staged by daily monitoring for vaginal mucus plugs. Occasional embryonic age corrections were made based on the developmental morphology of control littermate embryos.

Pregnant mice from timed matings received a combination of one 40 mg/kg intraperitoneal (IP) injection of Dox prepared in sterile PBS at the start of the dosing period and 100 µg/ml Dox added to the drinking water (replaced every 48 h) until the time of embryo harvesting.

EMT explant assays (Runyan and Markwald, 1983) used collagen gel matrices prepared as described (Xiong et al., 2012) using Opti-MEM (Invitrogen) with 2% fetal bovine serum (FBS; HyClone), 100 units/ml penicillin, 100 μg/ml streptomycin (HyClone) and 5 μg/ml doxycycline (MP Biomedicals). The medium was supplemented with 100 nM Wnt-C59 (Selleck Chemicals) where indicated. Intact outflow tracts or atrioventricular canals from E9.5 Tie2:Cre;R26R:rtTA;TRE:H2BGFP or Tie2:Cre;R26R:rtTA;TRE:H2BGFP;TRE:Dkk1 embryos (genotyped retroactively and, generally, from Dox-pretreated pregnant mice) were dissected in sterile PBS, filleted using sharp tweezers to expose the cushions and placed cushion-side down on the collagen gels. 24 h after explant culture in a dual gas 37°C incubator containing 5% carbon dioxide (CO₂) and 5% oxygen (O₂), bright-field and epifluorescent images used for cell counting were captured using a Nikon Eclipse TI inverted microscope. Explant co-cultures with wild-type OFTs followed (Chang et al., 2004) using the above procedures except images were captured after 48 h. The number of GFP⁺ cells was scored for individual explants. Data from multiple litters were combined and statistical significance determined by Student's two-tailed t-tests.

Cushion mesenchymal cells were prepared and cultured from E13.5 Tie2:Cre;Rosa:rtTA;TRE:H2BGFP embryos. Cells were treated with Wnt3A (R&D Systems), FGF1 and FGF2 (Peprotech), and/or the FGF receptor inhibitor PD-173074 (LC Laboratories). Proliferation rates of GFP-labeled EMT-derived cells in each treatment condition were determined by 5-ethynyl-2′-deoxyuridine (EdU) incorporation. Statistical differences were assayed by one-tailed Fisher's exact tests. Further details on the culture methods, drug treatments and EdU proliferation assays are presented in the supplementary Materials and Methods.

Immunostaining of paraffin sections of Nfatc1^Cre, Tie2:Cre, Axin2^lacZ, R26R:rtTA, TRE:H2BGFP and wild-type mice was performed at various embryonic stages as described in the supplementary Materials and Methods.

Western blots of whole embryo or whole heart lysates were probed with primary antibodies against Axin2, versican and tenascin-C as described in the supplementary Materials and Methods.

This project benefited from the early support of Dr Ching-Pin Chang, in whose lab the TRE:Dkk1 line was validated, and Dr Calvin Kuo, whose lab generated the TRE:Dkk1 founder transgenic lines. Dr Bin Zhou shared discussions and the Nfatc1^Cre line in advance of publication. Dr Steven Artandi provided Actin:rtTA mice. Dr Jill Helms shared Lef1-null tissue sections to validate the specificity of the anti-Lef1 antibody. We thank Andrew McKay, Brynn Akerberg, Astra Henner, Sarah Casper, Yujung Choi and Gene Ma for technical help and Stankunas lab members for feedback.