Bmp2 and Notch cooperate to pattern the embryonic endocardium

The early heart tube comprises primary myocardium (Moorman et al., 2003). As development proceeds, the myocardium becomes regionally specified into chamber and non-chamber tissue. The nascent chamber myocardium (ventricles and atria) is characterized by high proliferation rates (de Boer et al., 2012) and trabeculation (Sedmera et al., 2000). Non-chamber primary myocardium in the atrioventricular canal (AVC), outflow tract (OFT) and inner curvature is lined by extracellular matrix (ECM)-rich cushions and signals to the endocardium to activate the epithelial-mesenchyme transition (EMT) that gives rise to the mesenchyme of the valves and septa (Eisenberg and Markwald, 1995; Moorman et al., 2003). The primary myocardium in the AVC is characterized by expression of Bmp2 from embryonic day (E) 8.5 onward (Sugi et al., 2004; Ma et al., 2005; Singh et al., 2005), Tbx2 (Habets et al., 2002; Christoffels et al., 2004; Harrelson et al., 2004) and Tbx3 (Hoogaars et al., 2004). Bmp2 directly activates Tbx2 (Yamada et al., 2000; Ma et al., 2005).

Establishment of AVC myocardial identity and subsequent cushion formation are linked to EMT initiation and progression. Bmp2-deficient hearts show no AVC specification, fail to form endocardial cushions (Ma et al., 2005) and are unable to undergo EMT in explant cultures (Rivera-Feliciano and Tabin, 2006). Treatment of wild-type AVC explants with the Bmp antagonist Noggin represses EMT, as does the removal of the myocardial layer from untreated wild-type AVC explants. BMP2 is able to rescue the EMT in such cultures (Sugi et al., 2004) and to induce EMT in non-cushion-forming ventricular explants (Luna-Zurita et al., 2010), suggesting that Bmp2 is sufficient for EMT ex vivo.

Notch is a crucial signaling pathway during cardiac development, and its patterning role is highlighted by the specific expression of Notch elements in presumptive chamber and valve tissues (Del Monte et al., 2007). In early chamber endocardium, activation of the Notch1 receptor (N1ICD) is predominant in the endocardium at the base of the forming trabeculae, where Notch is essential for trabeculation (Grego-Bessa et al., 2007). In contrast, Notch is activated throughout the endocardium of presumptive valve territory (AVC and OFT); in these tissues, Notch promotes Snail transcription (Timmerman et al., 2004), and Bmp2-activated Smad leads to Snail activation and EMT (Luna-Zurita et al., 2010). Despite current knowledge of Bmp2-regulated AVC formation, EMT induction and its convergence with Notch, it remains unclear how Bmp2-mediated Smad affects endocardial Notch1 signaling.

Here, we examined the precise contribution of Bmp2 to AVC myocardial specification and its interplay with Notch to pattern the embryonic endocardium. We used a novel transgenic mouse model that enables conditional induction of Bmp2 and unraveled the transcriptional events that follow Bmp2 expression and lead to EMT induction, as well as the cooperation between Bmp and Notch in this process. EMT induction by Bmp2 is coordinated through the activation of a network of genes that regulate cell adhesion, intercellular communication, ECM deposition and proteasome-mediated protein degradation. EMT induction by Bmp2 does not take place in the absence of widespread Notch1 signaling. In fact, Bmp2 ectopic expression in chamber myocardium causes activation of the Notch pathway throughout the ventricular endocardium, which then acquires the ability to undergo EMT. This occurs through transcriptional activation of the Notch ligand Jag1 by Bmp2-pSmad1/5, via direct binding of Smad1 to the Jag1 promoter, as well as through the physical interaction of pSmad1/5 and N1ICD.

We generated transgenic mice conditionally overexpressing Bmp2 (Rosa26-Bmp2^GOF; hereafter Bmp2^GOF; see Materials and Methods; Fig. S1A,B). Bmp2^GOF mice were crossed with the pan-cardiac driver line Nkx2.5-Cre (Fig. 1A; Fig. S1C) (Stanley et al., 2002). Only 53% (8/15) and 8% (1/13) of Bmp2^GOF;Nkx2.5-Cre embryos examined survived to E10.5 and E11.5, respectively, with severe cardiac defects (Fig. S1D,E; Table S1A). At E9.5, Bmp2 expanded from its normal AVC myocardium expression to the myocardium and endocardium in transgenic embryos (Fig. 1B; Fig. S2A). This led to phosphorylation of the effectors Smad1/5 (Fig. 1C-E) and transcriptional activation of target genes (Tgfb2 and Smad6) in both ventricular myocardium and endocardium of Bmp2^GOF;Nkx2.5-Cre hearts (Fig. S2B,C).

To study the effect of Bmp2 overexpression on cardiac morphogenesis, we profiled E9.5 control and Bmp2^GOF;Nkx2.5-Cre hearts by RNA-seq. Transgenic hearts showed 5183 differentially expressed genes (DEGs), 2646 were upregulated and 2538 were downregulated (P<0.005) (Fig. 1F; Table S2, sheet one). The enrichment analysis of Gene Ontology (GO) terms revealed the over-representation of ECM, regulation of cell communication and signaling, cell adhesion, cancer pathways and cardiomyopathy terms for the upregulated genes, and RNA regulation, proteasome, cell cycle and proliferation terms for the downregulated genes (Fig. 1F; Table S2, sheets two and three).

To identify the DEGs with the closest hierarchical relation with respect to Bmp2, we generated a Bmp2-dependent network by integrating the 2000 DEGs with the lowest P-value with the STRING database (Szklarczyk et al., 2015). We isolated the DEGs with up to two degrees of neighborhood with Bmp2 and predicted a network of 57 nodes or genes (Fig. 1G). Among the direct interactors of Bmp2 (Fig. 1G, orange nodes), the Bmp receptor Bmpr1a, Smad effectors and Smurf2, a ubiquitin ligase that negatively regulates Smad proteins (Zhang et al., 2001), appeared upregulated, confirming the functional overactivation of Bmp2 (Fig. 1G). Interestingly, Smurf2 was connected to the majority of the second-degree interactors (Fig. 1G, yellow nodes), which were mostly downregulated and involved in proteasome regulation (Fig. 1G, bottom left). We confirmed the increased expression of both Smurf1 and Smurf2 in ventricular endocardium of transgenic hearts (Fig. S2D-F, black arrowheads).

The over-representation of ECM-related terms in the RNA-seq and the known role of Bmp2 in ECM regulation (Ma et al., 2005; Rivera-Feliciano and Tabin, 2006) prompted us to look into ECM deposition. Alcian Blue staining showed ECM accumulation (Fig. 1H,I and insets) and histological measurements revealed a significantly increased myocardium–endocardium distance in Bmp2^GOF;Nkx2.5-Cre ventricles (Fig. 1J). Increased expression of ECM-producing genes (Has2 and Vcan) and downregulation of ECM-degrading genes (Tmem2 and Hyal2) could explain the observed phenotype (Fig. 1K).

To identify the most affected processes within the high number of DEGs, we focused on the upregulated genes with the highest number of direct connections with other DEGs. GO analysis of this subgroup showed enrichment in cancer, cell adhesion and dilated cardiomyopathy terms (Fig. S2G; Table S2, sheet six). Cancer and adhesion terms are commonly related to EMT, cell migration and motility. We then compared our E9.5 Bmp2 GOF heart RNA-seq data with a E11.5 Bmp4 GOF mandibular tissue microarray (Bonilla-Claudio et al., 2012). We found 36 common upregulated genes, including known Bmp signaling targets, such as Id1, Msx2 and Tbx20. We also observed 95 genes that changed direction in the two data sets, which suggests tissue-specific transcriptional network regulation (Table S3). To obtain a more comprehensive comparison of Bmp regulation during early stages of development, we compared our current E9.5 Bmp2 GOF heart RNA-seq with an E14.5 Bmp2 GOF heart RNA-seq (Prados et al., 2018). This comparison (Table S4) highlighted several common upregulated and downregulated genes related to muscle, heart muscle and/or ventricular development (shown in dark red in Fig. S3), cardiac valve development (yellow in Fig. S3) and Bmp signaling (blue in Fig. S3).

We set out to characterize the effect of excessive myocardial Bmp2 on EMT in vivo. Hematoxylin & Eosin (H&E) staining of E9.5 Bmp2^GOF;Nkx2.5-Cre hearts revealed an expanded AVC (Fig. 2A,B, double arrows; Fig. 2C, measurements) filled with mesenchymal cells (Fig. 2A,B insets, arrowheads). Mesenchymal cells in transgenic embryos were also found throughout the ventricles (Fig. 2B inset, arrowhead), adjacent to the endocardium, which was localized in the center of the lumen instead of closely lining the myocardium (Fig. S4A-F).

The RNA-seq analysis of Bmp2^GOF;Nkx2.5-Cre hearts showed increased expression of cardiac EMT-related genes, such as Snail2/3, Sox9, Vim and Twist1 (Fig. 2D). Immunostaining or in situ hybridization in sections confirmed the expansion of Sox9, Snail1 and Twist1 to the ventricular endocardium in transgenic hearts (Fig. 2E-H; Fig. S4G,H). Thus, histological and expression data suggest that the transgenic ventricular endocardium molecularly resembles the endocardium of the AVC and, thus, that it acquired the ability to undergo EMT. To test this possibility, we used ex vivo explant cultures on collagen gels (Runyan and Markwald, 1983). Control and transgenic AVC endocardial cells were able to migrate and invade the gel (measured by their three-dimensional transformation index, 3D TI; Fig. 2I,J,M; Movies 1,2). Explanted control ventricles did not undergo EMT, but instead formed a continuous endothelial monolayer (Fig. 2K) (Luna-Zurita et al., 2010). By contrast, Bmp2^GOF;Nkx2.5-Cre ventricular endocardial cells transformed and invaded the gel, behaving like AVC endocardium (Fig. 2L,M; Movies 3,4). Transgenic ventricular endocardial cells migrated further and invaded deeper than their control counterparts, reaching similar levels to AVC endocardial cells (Fig. S4I,J). Interestingly, transgenic ventricular endocardial also lacked the expression of Irx5 (Fig. S4K,L), a known marker of ventricular endocardium (Christoffels et al., 2000; Bruneau et al., 2001), indicating the loss of ventricular endocardium identity.

The myocardium of transgenic ventricles was thinner and had small trabeculae (Fig. S5A), but a relatively normal gene expression profile, compared with control ventricles: it expressed chamber differentiation markers (Irx4, Myl2, Bmp10 and Nppa) and lacked AVC markers (Tbx2 and Tbx3) (Fig. S5B-K). Myocardial cell proliferation, assessed by BrdU incorporation, was reduced in Bmp2^GOF;Nkx2.5-Cre hearts, accounting for the thinner myocardium (Fig. S5L,M). These results were confirmed with earlier activation of Bmp2 throughout the mesoderm with the Mesp1-Cre driver (Fig. S6; Table S1B).

Notch1 activity is indispensable for EMT (Timmerman et al., 2004; Luna-Zurita et al., 2010). We next examined whether the expansion of EMT in Bmp2^GOF;Nkx2.5-Cre hearts was accompanied by expanded Notch1 activity. At E9.5, Notch1 pathway activation, detected by nuclear staining of N1ICD, occurred throughout the AVC endocardium (Fig. 3A, arrowhead) but was restricted to a subset of ventricular endocardial cells (Fig. 3A inset, arrowheads). N1ICD staining in Bmp2^GOF;Nkx2.5-Cre hearts was observed throughout the endocardium (Fig. 3B and inset), leading to a twofold increase in N1ICD protein levels (Fig. 3C,D). Our RNA-seq data showed increased expression of various components of the Notch pathway (Fig. 3E). The upregulation of Notch ligands by Bmp2 overexpression could explain the observed increase in Notch activity. Thus, we assessed the expression patterns of Dll4 and Jag1, the most relevant ligands at E9.5 (D'Amato et al., 2016; MacGrogan et al., 2016). Dll4, normally restricted to a subset of ventricular endocardial cells at this stage, was expanded throughout the endocardium in transgenic hearts (Fig. 3F, white arrowheads). Jag1, normally restricted to the myocardium in the ventricles, was found ectopically expressed in the ventricular endocardium of Bmp2^GOF;Nkx2.5-Cre hearts (Fig. 3G, white arrowheads). In contrast to Dll4, Jag1 did not appear upregulated in the RNA-seq, likely because of its predominant myocardial expression in normal E9.5 hearts, which, together with the myocardial thinning in transgenic hearts (Fig. 3G black arrowheads and quantification in Fig. S5M), masked any difference in gene expression levels. In situ hybridization of the Notch targets Hey1, Hey2 and Heyl confirmed the expanded Notch activity in the ventricular endocardium of transgenic hearts (Fig. 3H-J).

To establish whether Bmp2-dependent Notch activation is a prerequisite for EMT induction, we performed explant cultures, as described before, and blocked Notch signaling with the γ-secretase inhibitor RO4929097 (RO). Cell invasiveness in control and transgenic AVC explants (Fig. 3K,N) and in transgenic ventricular explants (Fig. 3L,O) was severely suppressed in the absence of Notch activity (Fig. 3M,P; Movies 5,6). Thus, Bmp2-induced ectopic EMT is mediated by Notch1.

To establish the source of Bmp2 responsible for the observed endocardial phenotype, we activated Bmp2 using myocardium- or endocardium-specific drivers (Fig. S7). Bmp2^GOF/+;cTnT-Cre embryos (myocardial Bmp2) recapitulated the Bmp2^GOF;Nkx2.5-Cre phenotype (ECM accumulation, expanded N1ICD and EMT; Fig. S8B,E,H,K,N,Q; Table S1C), whereas Bmp2^GOF/+;Tie2-Cre and Bmp2^GOF/GOF;Tie2-Cre embryos (endothelial Bmp2) appeared normal at E9.5 (Fig. S8C,F,I,L,O,R; Table S1D). Therefore, we concluded that induction of Notch1 and ectopic EMT in vivo is driven by myocardial Bmp2.

The ectopic endocardial expression of Jag1 in Bmp2^GOF;Nkx2.5-Cre ventricles (Fig. 3G) prompted us to study how this activation occurs. To test for a possible direct activation of Jag1 transcription by Bmp2-pSmad1, we performed chromatin immunoprecipitation (ChIP) experiments with an anti-Smad1 antibody on mouse embryonic ventricular endocardial cells (MEVECs) (D'Amato et al., 2016), treated or not with recombinant human BMP2. We assessed the binding of Smad1 to two Smad1/5-binding regions in the Jag1 locus (Morikawa et al., 2011): one in the promoter region (approximately 520 bp upstream of the transcriptional start site, TSS) and one within the second intron (Fig. S9). Quantitative PCR (qPCR) analysis of the immunoprecipitated DNA with primers flanking the Smad1-binding regions showed a significant enrichment of the promoter region after BMP2 stimulation in contrast to untreated cells (Fig. 4A). This enrichment was only observed after BMP2 treatment. We also observed a significant increase in the expression levels of Jag1 in this system (Fig. 4B), likely mediated by Smad1 binding to its promoter. Consistent with these data, activation of the Bmp pathway by transfection of the constitutively active Bmp receptor Alk3 (caAlk3) in endothelial cells, resulted in significant induction of a luciferase reporter containing the mouse Jag1 promoter (Fig. 4C) and only minimal induction of the Jag1 intronic region (Fig. 4D).

To understand how Bmp2 interacts with Notch in more depth, we used MEVEC co-cultured with OP9 cells expressing either GFP (control), or the Notch ligands Jag1 or Dll4 (Fig. S10A). OP9-Jag1 cells induced Notch activity, shown by Hey1, Hey2 and Heyl upregulation, in MEVEC (Fig. 5A, gray versus white bars). Addition of BMP2 in these co-cultures activated Bmp targets (Fig. S10B) and further increased the expression of Hey1 and Heyl (Fig. 5A, black/gray versus gray bars) and led to threefold increased N1ICD protein levels, as shown by western blot analysis (Fig. 5C, input and quantification). These findings suggest that Bmp2 enhances Notch pathway activity in vitro. Treatment of the OP9-GFP/MEVEC co-cultures with BMP2 also led to a significant increase in Hey1 and Heyl expression (Fig. 5A, black versus white bars), which suggests that Bmp2 activates the Notch pathway even in the absence of strong Notch signals. Bmp2-mediated activation of Jag1 expression, likely via direct binding of Smad1 to its promoter, could induce the Notch pathway in adjacent cells. Indeed, we observed the induction of Jag1 expression in BMP2-treated OP9-GFP/MEVEC co-cultures (Fig. 5B, black versus white bars). By contrast, Jag1 was not upregulated after Dll4-Notch induction (Fig. S10C; Fig. 5B, green bar) and no synergistic effect was observed after activation of both pathways (Fig. 5B, green/black bar). These findings suggest that Bmp2-mediated induction of Jag1 is independent of Notch, whereas activation of Dll4 by Bmp2 can only take place in the presence of Notch signals.

To explain the synergistic effect of Bmp and Notch signaling on gene expression, we tested for a potential physical interaction between N1ICD and pSmad1/5. After confirming the activation of N1ICD upon BMP2 stimulation (Fig. 5C), we immunoprecipitated (IP) OP9-Jag1/MEVEC (±BMP2) lysates with anti-Notch1 or anti-Smad1, and blotted with both antibodies. In BMP2-treated co-cultures, we detected pSmad1/5 in the Notch1-IP and N1ICD in the Smad1-IP (Fig. 5D,E), suggesting that the two proteins physically interact. We confirmed these results in BAEC, where we were able to co-IP pSmad1/5 and Myc-tagged N1ICD (Myc-N1ICDΔ^OP) (Milner et al., 1996) (running at 70 kDa; Fig. S11A,B). We also confirmed the ability of Bmp2 to induce Notch activity by performing luciferase assays on endothelial cells (BAEC). A Notch reporter containing ten consecutive CBF1/RBPJ-binding sites (McKenzie et al., 2005) was stimulated by caAlk3 in the absence and presence of N1ICD (Fig. 5F; Fig. S11C,D).

Given the low levels of the two proteins and the small size of the heart at E9.5, we were not able to perform this IP in vivo. However, we believe that MEVECs, isolated from E10.5 ventricles and stimulated in trans by OP9 cells (for Notch activation) or with recombinant BMP2 (for Bmp activation), are a physiologically relevant system. In vivo at E9.5, pSmad1/5 and N1ICD colocalized only minimally in ventricular endocardial cells (Fig. S12A, inset, white arrowhead). In transgenic hearts, the colocalization was more extensive (Fig. S12B, inset, white arrowheads), supporting a possible physical interaction between the proteins.

Together, our results show that Bmp2-pSmad1 is able to activate the transcription of the Notch ligand Jag1 by direct binding to its promoter, thus inducing Notch activity and target gene transcription. In cells where both pathways are active, Notch target gene expression is enhanced through the activation of Notch ligands and the interaction of the effectors pSmad1/5 and N1ICD.

Bmp2 is an essential signal that specifies the primitive AVC primary myocardium of the embryonic heart (Ma et al., 2005; Rivera-Feliciano and Tabin, 2006) versus the differentiated myocardium of the chambers. Myocardial Bmp2 expression drives a gene expression program that leads to formation of the valve primordium, in concert with endocardial-derived signals, such as Notch. Indeed, the differential distribution of N1ICD patterns the early valve and chamber endocardium, with widespread and relatively uniform expression in valve endocardium but restricted to the base of developing trabeculae in the ventricles (Timmerman et al., 2004; Del Monte et al., 2007; Grego-Bessa et al., 2007; MacGrogan et al., 2016). The combined activities of myocardial Bmp2 and endocardial Notch1 pattern the AVC and regulate the process of EMT (Luna-Zurita et al., 2010), which is crucial for valve development. Our study focused on the analysis of a novel mouse model conditionally overexpressing Bmp2, to identify global Bmp2-induced transcriptional changes associated with the regulation of valve EMT, and to shed light on the cooperation between Bmp2 and Notch1 signals.

Our results revealed the ability of pan-myocardial Bmp2 to specify ventricular endocardium as valve tissue, and to induce ectopic EMT in the ventricles. This is achieved via the transcriptional activation of several EMT drivers, including Snail1, Twist1 and Sox9, the expression of which is upregulated and expanded to the endocardium of the ventricles in transgenic embryos. Genes related to ECM regulation, cell adhesion, cell migration and intercellular signaling were also upregulated. Likewise, both Smurf1 and Smurf2, involved in the dissolution of tight junctions and in TGFβ-induced EMT (Ozdamar et al., 2005; Townsend et al., 2008; Chandhoke et al., 2016), were upregulated in Bmp2-overexpressing hearts. Thus, Smurf-mediated proteasome regulation of proteins (cytoskeletal, epithelial and junctional) might have a crucial role in the initiation of EMT under the control of Bmp2. The loss of expression of the chamber endocardium marker Irx5, together with the gain of the EMT drivers and effectors described above, suggest the conversion in transgenic embryos of ventricular endocardium into valve tissue.

Endocardial Notch1 pathway activation is a crucial step in EMT induction by Bmp2. Notch1 activity is sufficient to activate a pro-mesenchymal gene program through Snai11/2 induction, but completion of EMT and invasion of the cardiac jelly is achieved through Bmp2, which promotes Snail1 protein stabilization (Luna-Zurita et al., 2010). Bmp2-overexpressing hearts showed complete Notch1-dependent EMT in ex vivo ventricular cultures and in vivo within the ventricle. Transgenic ventricular endocardial cells migrated more efficiently than their control counterparts, but did not exceed the migration and invasion levels of AVC endocardial cells, either control or transgenic.

Interestingly, ectopic expression of Bmp2 in chamber myocardium of Bmp2^GOF;Nkx2.5-Cre embryos was unable to activate a full AVC myocardial program in the chambers, although this was not necessary for endocardial EMT to occur. Overexpression of Tbx2 (or Tbx3) in the heart tube represses ventricular myocardium genes (Christoffels et al., 2004; Hoogaars et al., 2007; Shirai et al., 2009; Singh et al., 2012). Bmp2 is essential for the activation of Tbx2 through pSmad1/5 transcriptional induction, which is in turn regulated by Tbx20 (Harrelson et al., 2004; Shirai et al., 2009; Singh and Kispert, 2010). High Tbx20 levels induced by Bmp2 could prevent the excessive pSmad1/5 from transcriptionally activating Tbx2, allowing for chamber myocardium genes to be normally expressed.

Our initial observation that ectopic Bmp2 profoundly affected Notch1 activity in vivo prompted us to examine how this activation occurs. We showed that Notch1 activity was responsive to Bmp2 signals in endocardial and endothelial cells in vitro: N1ICD was increased, Hey genes were upregulated, and a Notch reporter was stimulated after Bmp pathway activation. Here, we propose that this activation occurs in two distinct ways, depending predominantly on the endogenous Notch levels: protein interaction and transcriptional induction.

We showed in two different cellular models that pSmad1/5 and N1ICD physically interact. Smad1 and Smad3 can bind N1ICD, recruit p300 and stabilize the Notch transcriptional complex, along with P/CAF, in specific promoter regions, thus reinforcing pathway activity in myoblasts, endothelial and neuroepithelial cells (Blokzijl et al., 2003; Takizawa et al., 2003; Itoh et al., 2004; Moya et al., 2012). In epithelial cells, the common Bmp mediator Smad4 is also important for the synergistic activation of a Notch reporter (Itoh et al., 2004). Binding of pSmad1/5 to N1ICD in endocardial cells could act in a similar manner to stabilize the Notch complex. However, this protein interaction likely takes place only in the presence of high pSmad1/5 and N1ICD levels within the same cell. In N1ICD-negative or N1ICD-low cells, pSmad1/5 might trigger Notch pathway induction via the transcriptional activation of its ligands.

We also showed that Bmp2 alone transcriptionally activates the Notch ligand Jag1 in endocardial cells, by direct binding of Smad1 to its promoter, as has been shown in other systems (Morikawa et al., 2011). In turn, Jag1 binds to Notch receptors in adjacent cells, activating the pathway and inducing Hey gene transcription. Once the Notch pathway is active, BMP2 further enhances the expression of Dll4, Hey1 and Heyl.

We suggest that, in E9.5 wild-type hearts, the specification of AVC endocardium results from the interplay between myocardial Bmp2 and endocardial Notch1 signaling (Fig. 6). pSmad1/5 transcriptionally activates Notch ligands in AVC endocardium, leading to widespread Notch1 activity in this tissue. This allows for physical interaction between the abundant pSmad1/5 and N1ICD, which further enhances Notch1 activity. The combined effect of both pathways triggered the transcription in the AVC of genes inducing EMT (Fig. 6A,C). In the ventricles, low Bmp signaling and restricted endocardial Notch activity prevented the physical interaction of the effectors, the endocardium showed features of chamber tissue, and EMT did not occur (Fig. 6A,D). Ectopic Bmp2 expression in the ventricles increased pSmad1/5 levels, resulting in transcriptional activation of Notch ligands, expansion of Notch pathway activity, and cooperation between the two pathways for EMT induction and valve specification in both AVC and chamber tissue (Fig. 6B,E).

Animal studies were approved by the CNIC Animal Experimentation Ethics Committee and the Community of Madrid (Ref. PROEX 118/15). All animal procedures conformed to EU Directive 2010/63EU and Recommendation 2007/526/EC regarding the protection of animals used for experimental and other scientific purposes, enforced in Spanish law under Real Decreto 1201/2005. The following mouse strains were used: Bmp2^GOF, Nkx2.5-Cre (Stanley et al., 2002), cardiac TnT-Cre (Jiao et al., 2003), Tie2-Cre (Kisanuki et al., 2001) and Mesp1-Cre (Saga et al., 1999). To generate Bmp2^GOF mice, a cDNA encoding Bmp2 was cloned into a modified version of the pROSA26-1 plasmid, preceded by a loxP-flanked ‘Neo^R-STOP’ cassette and followed by an IRES-EGFP cassette. Gene targeting, performed in G4 embryonic stem cells, was confirmed by Southern blotting with 5′ and 3′ probes. Mice were generated by injection of targeted cells into B6CRL blastocysts to generate chimeras that were then analyzed for germline transmission. The selected animals were backcrossed to the C57BL/6 background. All reported controls are Cre driver heterozygotes.

H&E, Alcian Blue staining, X-gal staining and in situ hybridization on paraffin sections and in whole embryos were performed as described (de la Pompa et al., 1997). The primers used were: Smurf1 F, 5´-GGAAACCTGTGGCTTTGCAG-3´; R, 5´-CCATCACACCTGCTTTGTGC-3´; Smurf2 F, 5´-CACGGTTGCTCCAGTTTGTG-3´; R, 5´-CCGACTGCTGGAGTTGTCTT-3´. Details of the other probes used can be found in D'Amato et al. (2016), MacGrogan et al. (2016), Christoffels et al. (2000) and Bruneau et al. (2001).

Paraffin sections (7 μm thick) were incubated overnight with primary antibodies, followed by 1 h incubation with the appropriate secondary antibody conjugated to a fluorescent dye. Staining for N1ICD, Jag1 and phospho-Smad1/5 was performed using tyramide signal amplification (Del Monte et al., 2007). For same-species double staining (pSmad1/5 and N1ICD), two sequential assays were performed combining two unlabeled rabbit antibodies in separate detection sequences. Cross-reactions were prevented by extensive blocking with excess unconjugated Fab anti-rabbit IgG (1 h) and 5 min heating in citrate buffer between the two assays. Experiments omitting the first or second primary antibody were used routinely to control for unwanted background staining. Antibodies were used against the following targets: phospho-Smad1/5 (Cell Signaling, 9516, 1:100), Sox9 (Santa Cruz, sc-20095, 1:100), BrdU (AbCam, ab6326, 1:100), Isolectin GS-IB4-Alexa Fluor 647 (Molecular Probes, mp132450, 1:200), cleaved Notch1 (N1ICD, Cell Signaling, 4147, 1:100), α-smooth muscle actin-Cy3 (Sigma, C6198, 1:200) and Snail1 (1:50; Francí et al., 2006).

Cell proliferation was analyzed at E9.5 after a 2 h BrdU pulse by detecting incorporated BrdU on paraffin sections. The number of positive nuclei, in both the myocardium (isolectin-B4-negative cells) and endocardium (isolectin-B4-positive cells), was divided by the total number of nuclei counted on four sections (at different levels) in three embryos per genotype. pSmad1/5-positive cells were counted similarly. Images were processed with ImageJ (image processing and analysis in Java). Data are presented as mean±s.d.

Measurements were performed with ImageJ on H&E-stained paraffin sections using morphological criteria. The length of the AVC was measured by drawing a line from the boundary between the atrium and AVC to the boundary between the ventricle and AVC (determined by the presence of trabeculae). The area occupied by the AVC was measured by delimiting the AVC using the aforementioned criteria, applying the ‘area’ tool and normalizing against the area of the heart at the four-chamber level. The length of the trabeculae and the distance between endocardium and myocardium were calculated by taking several measurements from the compact myocardium to either the tip of the trabeculae (six trabeculae per section when available) or the endocardium (six different areas throughout the ventricle). Three sections (at different levels) of four to seven embryos per genotype were measured (each data point represents an embryo). Data are presented as mean±s.d.

Explant cultures were performed as previously described (Runyan and Markwald, 1983; Luna-Zurita et al., 2010). For treatments with the γ-secretase inhibitor RO (Munch et al., 2013), collagen gels were pre-conditioned in explant medium containing 20 μM RO or vehicle (DMSO). The myocardium was removed from RO-treated explants for better z-axis visualization. Data are presented as mean±s.d.

Explants were fixed and stained with phalloidin-Alexa Fluor 488 (Invitrogen, A12379) and anti-α-smooth muscle actin-Cy3 (1:100; Sigma, C6198) to detect the actin cytoskeleton and mesenchymal cells, respectively. Explants were then mounted on excavated slides in Vectashield medium containing DAPI.

The 3D TI of AVC and ventricular explants is the ratio of the number of cells invading the collagen gel to the total number of cells counted by DAPI staining, excluding the myocardium. The average distance and depth of invasion were measured manually in ImageJ on x-y and x-z axes images. Six independent measurements were performed on each explant. A total of 10-14 explants were examined per condition. Data are presented as mean±s.d.

Confocal images of stained explants and tissue sections were acquired with a Nikon A1R laser-scanning confocal microscope. Images of stained explants were collected as z-stacks. z-projections and lateral sections were assembled using ImageJ.

Immortalized MEVECs, derived from E10.5 ventricles, were cultured as described (D'Amato et al., 2016). MEVECs were co-cultured for 6 h with GFP-, Jag1- or Dll4-expressing OP9 cells at 80% confluence in DMEM supplemented with 10% fetal bovine serum (FBS) and antibiotics. Overnight serum-free incubation was followed by a 5-h treatment with 20 ng/ml human recombinant BMP2. All cells were then washed in cold PBS and harvested for either immunoprecipitation or RNA extraction. Alternatively, MEVECs were incubated overnight in serum-free medium and then treated for 5 h with BMP2, as previously described. Material was collected for ChIP assay, as described below. Bovine aortic endothelial cells (BAECs) were cultured on 0.1% gelatin-coated plates in complete DMEM supplemented with 10% FBS and antibiotics, and were used for luciferase assays at 95% confluence.

Total RNA from cells was extracted with Trizol (Invitrogen). cDNA was synthesized from 1 μg of total RNA using the SuperScript III First Strand kit (Invitrogen). qPCR was performed with the Power SYBR Green Master Mix (Applied Biosystems) and KiCqStart Primers (Sigma-Aldrich). Tissue samples from seven embryos of each genotype were pooled; three pools were used for the analysis. Data are presented as mean±s.d.

The Evolutionary Conserved Region platform (ECR; http://ecrbrowser.dcode.org) was used to identify the mouse orthologs of the two previously identified SMAD1-binding regions in the human JAG1 locus, within the second intron and the promoter, approximately 520 bp upstream of the transcription start site (TSS) (Morikawa et al., 2011). Genomic DNA from E9.5 mouse embryos was used as a template to amplify 1162 bp (intronic) and 761 bp (promoter), containing the regions of interest (Fig. S9). The intronic product was cloned into the KpnI/XhoI sites and the promoter product into the KpnI/HindIII sites of the pGL3 luciferase plasmid. The primer sequences were: 5′-ATGCGGTACCCATCTGGCAGTAAGGTGGCT-3′ (intronic, forward, KpnI site underlined), 5′-ATCGCTCGAGTGAACTGGACGGTTGTGCTT-3′ (intronic, reverse, XhoI site underlined), 5′-ATGCGGTACCAAGGAACCTGGAAGGACCGT-3′ (promoter, forward, KpnI site underlined), and 5′-ATCGAAGCTTCCAGGAGTGTCGCTGCTAAT-3′ (promoter, reverse, HindIII site underlined).

BAECs were transiently transfected with the Notch reporter pGL3-10xCBF1-Luciferase (McKenzie et al., 2005) or the Bmp reporter pBRE-Luciferase (Korchynskyi and ten Dijke, 2002) or the mouse Jag1 genomic luciferase constructs at a 5:1 ratio with pGL3-Renilla using Lipofectamine LTX (Life Technologies). Notch stimulation was achieved by transient co-transfection of a Myc-tagged N1ICD-expressing plasmid (N1ICDΔ^OP) (Milner et al., 1996) and Bmp stimulation by co-transfection of a pcDNA3 plasmid expressing constitutively active Alk3 (caAlk3) (ten Dijke et al., 1993), kindly provided by P. ten Dijke (Leiden University, The Netherlands). Fetal bovine serum was added to the medium 4 h after transfection, to a final concentration of 5%. After 20 h of incubation, luciferase activity was measured with the Dual Luciferase Assay System (Promega). The firefly/Renilla luciferase ratios are presented as mean±s.d. Experiments were performed at least three times, in quadruplicate for each experiment. Differences were considered statistically significant at P<0.05, as assessed by one-way ANOVA.

MEVECs (untreated and BMP2-treated) were fixed in 1% formaldehyde for 5 min at room temperature and quenched by glycine 125 mM for 5 min at room temperature. After three washes in cold PBS, cells were harvested and snap-frozen in liquid nitrogen. Briefly, pellets from 2×10⁷ cells were re-suspended and incubated in 1 ml of Lysis Buffer 1 [50 mM HEPES-KOH (pH 7.5), 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100] and 1 ml of Lysis Buffer 2 (200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 10 mM Tris pH 8) consecutively. The resulting pellet was re-suspended in Lysis Buffer 3 (1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, in 1×PBS) and chromatin was sheared by sonication using a Bioruptor ultrasonicator (Diagenode). Sheared chromatin was split and immunoprecipitated with 10 μl of anti-Smad1 (Cell Signaling, 6944) or rabbit IgG (Santa Cruz Biotechnology, sc-2027).

Equal amounts of input and IgG/Smad1 immunoprecipitated DNA were used for qPCR analysis. The primer sequences were: 5′-GCACCATGAAGCTGGAAGAATCT-3′ (negative, forward), 5′-TACCCAAAGGAGCATAGCGTT-3′ (negative, reverse), 5′-TGGTCTAAATCGGGCCACTAAC-3′ (positive, Smad7, forward), 5′-AGCAGGGGGAAATCGGAAGAG-3′ (positive, Smad7, reverse), 5′-GCTGGGAGCTTCCTTTTCCT-3′ (Jag1 intronic, forward), 5′-ATGGCGGTTTGTTTGTGTCG-3′ (Jag1 intronic, reverse), 5′-CAGACTCTGCTGGGAACTTTG-3′ (Jag1 promoter, forward) and 5′-GGACATGGATTTGGGAAGG-3′ (Jag1 promoter, reverse). For data quantification, after normalizing to input, Smad1/IgG fold-enrichment was calculated. Differences were considered statistically significant at P<0.05, as assessed by one-way ANOVA for multiple comparisons. Data are represented as mean-fold enrichment±s.d. of at least three independent experiments.

Cells were harvested from 100 mm culture dishes in IONIC buffer (50 mM TrisHCl pH 8.0, 150 mM NaCl, 0.5% NP40, 0.5% sodium deoxycholate, 0.05% SDS) containing protease, phosphatase and proteasome inhibitors (MG-132). After crosslinking of the antibody of interest to Protein G Dynabeads (Thermo Fisher Scientific, 10003D) with dimethyl pimelimidate, the protein lysates were centrifuged and immunoprecipitated overnight at 4°C. The antibody-bound proteins were then eluted from the beads in 0.5 M glycine pH 2.5. Then, TrisHCl 1M (pH 8.8) was added at 10% of the sample volume to restore the pH. The antibodies used for immunoprecipitation were anti-c-Myc (5 μg; Invitrogen, 13-2500), anti-Notch1 (1:50; Cell Signaling, 3608) and anti-Smad1 (1:100; Cell Signaling, 6944). The Myc antibody was used for the detection and precipitation of the Myc-tagged form of N1ICD (N1ICDΔ^OP) (Milner et al., 1996).

Three pools of six control and six Bmp2^GOF;Nkx2.5-Cre E9.5 hearts were lysed in RIPA buffer (TrisHCl 50 mM pH 8.0, NaCl 150 mM, NP40 1%, sodium deoxycholate 0.5%, SDS 0.1%) containing protease and phosphatase inhibitors at 4°C for 30 min. After centrifugation at 16000 g for 20 min at 4ºC, the supernatants were denatured with Laemmli buffer at 95°C for 5 min. Equal amounts of input (25 μg) and immunoprecipitated samples or 60 μg of cardiac lysates were separated by SDS-PAGE and transferred to PVDF membranes (Millipore) under standard conditions. Blots were probed with antibodies to phospho-Smad1/5 (1:1000; Cell Signaling, 9516), Notch1 (1:1000; Cell Signaling, 3608), c-Myc (1:2000; Invitrogen, 13-2500), PCNA (1:5000; BD Biosciences, 610665) and alpha-tubulin (1:5000; Sigma, T8203). Bound HRP-conjugated secondary antibodies were detected by ECL (GE Healthcare). Band intensities were quantified by densitometric analysis using ImageJ software.

Total RNA was isolated at E9.5 from whole hearts of control and Bmp2^GOF;Nkx2.5-Cre embryos (four pools of six hearts per genotype, 24 embryos in total per genotype) using the Arcturus PicoPure RNA isolation kit (AB, 12204). The RNA-seq was performed in the CNIC Genomics Unit: 200 ng of RNA were used to generate barcoded RNA-seq libraries using the NEBNext Ultra RNA Library preparation kit (New England Biolabs). Briefly, polyA+ RNA was purified using poly-T oligo-attached magnetic beads followed by fragmentation and then first and second cDNA strand synthesis. Next, cDNA 3′ ends were adenylated and the adapters were ligated, followed by PCR library amplification. Finally, the size of the libraries was checked using the Agilent 2100 Bioanalyzer DNA 1000 chip and their concentration was determined using the Qubit fluorometer (Life Technologies). Libraries were sequenced on a HiSeq2500 (Illumina) to generate 60-bp single reads. FastQ files for each sample were obtained using CASAVA v1.8 software (Illumina). Sequencing reads were preprocessed by means of a pipeline that used FastQC (www.bioinformatics.babraham.ac.uk/projects/fastqc/) to assess read quality, and Cutadapt 1.7.1 to trim sequencing reads, eliminating Illumina adaptor remains and discarding reads that were shorter than 30 bp. The resulting reads were mapped against the mouse transcriptome (GRCm38, release 76; aug2014 archive) and quantified using RSEM v1.2.20 (Li and Dewey, 2011). Data were then processed with a differential expression analysis pipeline that used the Bioconductor package limma (Ritchie et al., 2015) for normalization and differential expression testing.

The 2000 DEGs with the lowest P values identified by our RNA-seq were used as input node data into the STRING 10.0 (http://string-db.org/) database (Szklarczyk et al., 2015). To obtain more stringent interactions, the Textmining interaction source was excluded, and the minimum required interaction score considered was 0.7 (high confidence). Resultant networks were then replotted using Cytoscape 3.4.0 (www.cytoscape.org/) (Shannon et al., 2003) with the force-directed layout setting. Hubs were defined as the upregulated nodes with the most direct neighbors of the 2000 DEGs.

GO and KEGG Pathway enrichment analyses of DEGs were performed using GO-Elite (www.genmapp.org/go_elite/; 1.2.5, EnsMart77Plus database version) (Zambon et al., 2012). For the analysis of the RNA-seq results, we used as background the 13,637 genes considered to be expressed. For the comparison between the RNA-seq from this report and RNA-seq from Prados et al. (2018), we used as background the 12,435 genes considered to be expressed in both data sets. z-score of the terms in the different categories (Biological process, Cellular component, Molecular function and KEGG pathways) were represented in heatmaps using GraphPad Prism 7.

Statistical significance was assessed using Student’s t-test, unless otherwise stated (*P<0.05, **P<0.01, ***P<0.001).

We thank V. Bou and A. Galicia for animal husbandry, S. Bartlett (CNIC) for English editing, and Fátima Sanchez for statistical support. We also thank the Genomics, Bioinformatics, Pluripotent Cell Technology and Transgenesis units at the CNIC.