Coronary arterial development is regulated by a Dll4-Jag1-EphrinB2 signaling cascade

Coronaries are essential for myocardial growth and heart function. Notch is crucial for mouse embryonic angiogenesis, but its role in coronary development remains uncertain. We show Jag1, Dll4 and activated Notch1 receptor expression in sinus venosus (SV) endocardium. Endocardial Jag1 removal blocks SV capillary sprouting, while Dll4 inactivation stimulates excessive capillary growth, suggesting that ligand antagonism regulates coronary primary plexus formation. Later endothelial ligand removal, or forced expression of Dll4 or the glycosyltransferase Mfng, blocks coronary plexus remodeling, arterial differentiation, and perivascular cell maturation. Endocardial deletion of Efnb2 phenocopies the coronary arterial defects of Notch mutants. Angiogenic rescue experiments in ventricular explants, or in primary human endothelial cells, indicate that EphrinB2 is a critical effector of antagonistic Dll4 and Jag1 functions in arterial morphogenesis. Thus, coronary arterial precursors are specified in the SV prior to primary coronary plexus formation and subsequent arterial differentiation depends on a Dll4-Jag1-EphrinB2 signaling cascade.


Introduction
Coronary artery disease leading to cardiac muscle ischemia is the major cause of morbidity and death worldwide (Sanchis-Gomar et al., 2016). Deciphering the molecular pathways driving progenitor cell deployment during coronary angiogenesis could inspire cell-based solutions for revascularization following ischemic heart disease. The coronary endothelium in mouse derives from at least two complementary progenitor sources (Tian et al., 2015) that may share a common developmental origin (Zhang et al., 2016a;Zhang et al., 2016b). The sinus venosus (SV) commits progenitors to arteries and veins of the outer myocardial wall (Red-Horse et al., 2010;Tian et al., 2013), and the endocardium contributes to arteries of the inner myocardial wall and septum (Red-Horse et al., 2010;Wu et al., 2012;Tian et al., 2013). Regardless of origin, the endothelial precursors invest the myocardial wall along stereotyped routes and eventually interlink in a highly coordinated fashion. Subsequently, discrete components of the primitive plexus are remodeled into arteries and veins, and stabilized through mural cell investment and smooth muscle cell differentiation (Udan et al., 2013). Arterial-venous specification of endothelial progenitors is genetically pre-determined (Swift and Weinstein, 2009), whereas arterial differentiation and patterning depend on environmental cues, such as blood flow and hypoxia-dependent proangiogenic signals (le Noble et al., 2005;Jones et al., 2006;Fish and Wythe, 2015). Vascular endothelial growth factor (VEGF) binds endothelial receptors and drives the expansion of the blood vessel network as a response to hypoxia (Dor et al., 2001;Liao and Johnson, 2007;Krock et al., 2011).
The Notch signaling pathway is involved in angiogenesis in the mouse embryo and in the postnatal retina. In this processes, the ligand Dll4 is upregulated by VEGF, leading to Notch activation in adjacent endothelial cells (ECs), vessel growth attenuation, and maintenance of vascular integrity (Blanco and Gerhardt, 2013). In contrast, the ligand Jag1 has a proangiogenic Dll4-Notch-inhibitory function, suggesting that the overall response of ECs to VEGF is mediated by the opposing roles of Dll4 and Jag1 (Benedito et al., 2009). Dll4-Notch1 signaling is strengthened in the presence of the glycosyltransferase Mfng (Benedito et al., 2009;D'Amato et al., 2016).
Our understanding of how coronary vessels originate, are patterned, and integrate with the systemic circulation to become functional is still limited. Coronary arteries are distinct from peripheral arteries in that they originate from SV and ventricular endocardium, which is a specialized endothelium that lines the myocardium. Moreover, SV endothelium has a venous identity, as opposed to the retina vascular bed, which has no pre-determined venous identity. Given these differences of developmental context, it is essential to evaluate the role of Notch in coronary arterial development, and importantly, the implications for heart development and repair.
Several components of the Notch pathway have been examined in the context of coronary artery formation. Inactivation of the Notch modifier Pofut1 results in excessive coronary angiogenic cell proliferation and plexus formation (Wang et al., 2017), while endothelial inactivation of Adam10, required for Notch signaling activation, leads to defective coronary arterial differentiation (Farber et al., 2019). Transcriptomics has shown that pre-artery cells appear in the immature coronary vessel plexus before coronary blood flow onset, and express Notch genes, including Dll4 (Su et al., 2018). Here, we examine the early and late requirements of Notch ligands Jag1 and Dll4, and their downstream effector EphrinB2, for coronary arterial development.

Jag1 and Dll4 are expressed in SV endocardium and coronary vessels endothelium
We examined SV and coronary vessels for the expression of Jag1 and Dll4. At embryonic day 11.5 (E11.5), Jag1 was detected in SV ECs and in ECs extending into the right atrium ( Figure 1A). Dll4 was also expressed in ECs emanating from the SV and in the endocardium lining the right atrium ( Figure 1A). These results suggest that either ligand could potentially activate Notch1 in the SV endocardium ( Figure 1A). At E12.5, N1ICD was detected in endomucin (Emcn)-positive ECs in subepicardial capillaries emerging from the SV ( Figure 1A). At E13.5, Jag1 and Dll4, were expressed in ECs of the developing coronary arteries (intramyocardial vessels; Figure 1-figure supplement 1A). Dll4 and Mfng were also expressed in prospective veins (subepicardial vessels; Figure 1-figure supplement 1A). At E15.5, Jag1, Dll4, and Mfng were all expressed in arterial ECs, whereas Mfng was still found in subepicardial vessels ( Figure 1-figure supplement 1B). Thus, Jag1 and Dll4 expression is found in discrete ECs in SV endothelium, abates in sub-epicardial veins and becomes restricted to intramyocardial coronary arteries at later developmental stages.

Nfatc1-positive progenitors give rise to the majority of subepicardial vessels
To inactivate Notch ligands in SV progenitors we used the Nfatc1-Cre driver line (Wu et al., 2012). To confirm the SV and endocardial specificity of this line, we crossed it with the Rosa26-LacZ reporter line (Soriano, 1999). X-gal-staining of heart sections of E11.5 embryos identified patchy LacZ expression in SV endothelium (Figure 1-figure supplement 2A). ß-gal staining was consistent with Nfatc1 protein nuclear localization in a subset of ECs lining the SV (Figure 1-figure supplement 2A). Uniform ß-gal staining was detected in ventricular endocardium and cushion mesenchyme derived from endocardial cells (Figure 1-figure supplement 2A). At E12.5, co-labelling with an anti-Pecam1 antibody revealed ß-gal-positive staining in 60% of Pecam1-positive subepicardial vessels in the right ventricle and about 50% in the left ventricle (Figure 1-figure supplement 2B-Source data 1, sheet 1). Tracking the expression of Nfatc1-Cre-driven red fluorescent protein (RFP) and the endothelial-specific nuclear protein Erg at E12.5 indicated that about 80% of nuclei in the Figure 1. Endocardial Jag1 or Dll4 inactivation disrupts coronary plexus formation. (A) Jag1, Dll4, and N1ICD immunostaining (red) in E11.5 control hearts, sagittal views. Magnified views of boxed areas show details of sinus venosus (sv, arrowheads) and right atrium (ra, arrow). Whole-mount dorsal view of immunostainings for N1ICD (red) and Emcn (green) in E12.5 control heart. Magnified views show detail of sub-epicardial endothelium. Nuclei are counterstained with Dapi (blue). (B) Whole-mount dorsal view of immunostaining for Emcn (green) in E12.5 control, Jag1 flox ;Nfatc1-Cre, and Dll4 flox ; Figure 1 continued on next page endothelial network were Nfatc1-positive and Erg-positive (Figure 1-figure supplement 2B-Source data 1, sheet 1). Thus, SV-derived Nfatc1-positive progenitors give rise to 50-80% of subepicardial vessels in the ventricular wall, consistent with previous reports (Chen et al., 2014;Cavallero et al., 2015;Zhang et al., 2016a).
To trace the fate of Nfatc1-positive cells relative to Notch activity, we crossed Nfatc1-Cre; Rosa26-RFP mice with the Notch reporter line CBF:H2B-Venus. At E11.5, a subset of nuclear-stained RFP ECs extending from the SV into the right ventricle were co-labelled with CBF:H2B-Venus (Figure 1-figure supplement 2D). At E12.5, some RFP-labelled capillaries on the dorsal side of the heart were co-labeled with CBF:H2B-Venus while others were labelled with CBF:H2B-Venus alone ( Figure 1-figure supplement 2E), indicating that Notch signaling activity is present in both Nfatc1positive and Nfatc1-negative populations of SV-derived ECs.
Opposing roles of Jag1 and Dll4 in coronary plexus formation from SV Jag1 inactivation with the Nfatc1-Cre driver line, specific of SV and endocardium, resulted in death at E13.5 (Supplementary file 1). Whole-mount Endomucin (Emcn) staining at E12.5 revealed a wellformed vascular network covering the dorsal aspect of control hearts ( Figure 1B, E-figure supplement 3A-Source data 1, sheet 2), whereas the vascular network in Jag1 flox ;Nfatc1-Cre mutants was poorly developed, and exhibited numerous capillary malformations ( Figure 1B, E-figure supplement 3B-Source data 1, sheet 2) with decreased endothelial branching ( Figure 1B,E-Source data 1, sheet 2). N1ICD expression was also more prominent ( Figure 1C) and EC proliferation reduced 60% relative to control ( Figure 1D,E-Source data 1, sheet 2). Ventricular wall thickness in endocardial Jag1 mutants was reduced by 50-60% relative to controls (Figure 1-figure supplement 4A-Source data 1, sheet 3). Thus, endocardial Jag1 deletion causes the growth arrest of the primitive coronary plexus.
In situ hybridization (ISH) showed reduced expression of HeyL and the Notch target Efnb2 in subepicardial vessels ( Figure 2B). Fabp4, a member of the fatty-acid-binding protein family, was upregulated ( Figure 2A). Fabp4 is a DLL4-NOTCH target downstream of VEGF and FOXO1 in human EC (Harjes et al., 2014), required for EC growth and branching (Elmasri et al., 2009). Fapb4 expression was found exclusively in the atrio-ventricular groove in Jag1 flox ;Nfatc1-Cre hearts ( Figure 2B), but extended sub-epicardially into the base of the heart in Dll4 flox ;Nfatc1-Cre hearts ( Figure 2B). Vegfa expression was not globally affected in mutant hearts ( Figure 2B), despite being upregulated in the RNA-seq ( Figure 2A). Cell cycle-associated genes such as Cdkn1b/p27, a negative regulator of cell proliferation, were also upregulated in both genotypes (Figure 2A), and p27 nuclear staining was increased twofold in compact myocardium ( Figure 2C,D-Source data 1, sheet 4), indicating decreased cellular proliferation. Connexin 40 (Cx40) and Hey2, which label trabecular and compact myocardium respectively, showed no alteration in their expression domains by ISH ( Figure 2B We examined E12.5 Jag 1flox ;Nfatc1-Cre and Dll4 flox ;Nfatc1-Cre mutants for evidence of hypoxia given that the gene signatures in the RNA-seq analysis suggested an ongoing hypoxic/metabolic stress response. The hypoxic response might also explain the defect of ventricular wall growth. We performed immunohistochemical detection of pimonidazole on E12.5 Jag1f lox ;Nfatc1-Cre and These results indicate that Jag1 flox ;Nfatc1-Cre and Dll4 flox ;Nfatc1-Cre mutant hearts are not overtly hypoxic at E12.5, suggesting that the hypoxic/metabolic stress gene signatures may be due to a cell autonomous defect of endocardial/endothelial cells. Defective coronary remodeling and maturation in endothelial Jag1 or Dll4 mutants We next examined the requirements of endothelial Jag1 and Dll4 for coronary vessel remodeling and maturation. To circumvent the early lethality of Jag1 flox -or Dll4 flox ;Nfatc1-Cre mutants, we crossed Jag1 flox and Dll4 flox mice with the vascular endothelium-specific Pdgfb-iCre ERT2 transgenic mice (Wang et al., 2010) to obtain the corresponding tamoxifen-inducible lines. Tamoxifen-induced Jag1 deletion at E12.5 resulted in 68% reduction in Jag1 expression (Figure 3-figure supplement 1A,C-Source data 1, sheet 5) and a complete absence of arteries at E15.5, whereas the veins appeared unaffected ( Figure 3A,C-Source data 1, sheet 6). We measured NOTCH pathway activity by carrying out a N1ICD staining that showed a 50% increase in endothelial N1ICD ( Figure 3B, C-Source data 1, sheet 6), consistent with Jag1 acting as an inhibitory Notch ligand. Next, we examined perivascular coverage of the Jag1 flox ;Pdgfb-iCre ERT2 endothelial coronary tree. We used a-smooth muscle actin (aSMA) and Notch3 to measure the extent of coronary vessel smooth muscle cell differentiation and pericyte coverage, respectively (Volz et al., 2015). We found that the proportion of aSMAand Notch3-positive cells was significantly reduced in E15.5 Jag1 flox ;Pdgfb-iCre ERT2 coronary arteries ( Figure 3B,C-Source data 1, sheet 6) reflecting the lack of differentiation of perivascular cells. Although aSMA is a commonly used marker of vascular smooth muscle cells, it is expressed more broadly in mesenchymal cells and cardiomyocytes throughout the embryonic heart prior to E16.5. To evaluate smooth muscle differentiation more specifically in E15.5 hearts, we used SM22a (Figure 3-figure supplement 2A,B). After co-staining with Notch3, we found that the proportion of SM22a and Notch3-positive cells in the Jag1 flox ;Pdgfb-iCre ERT2 mutants was significantly reduced (Figure 3-figure supplement 2A,B,E-Source data 1, sheet 7), confirming that pericytes fail to properly differentiate into smooth muscle.

Forced endothelial Dll4 expression disrupts coronary vascular remodeling
To further characterize the role of Notch in coronary development, we generated a transgenic line (Dll4 GOF ) bearing a Rosa26-CAG-floxNeoSTOPflox-Dll4-6xMycTag expression cassette (see Materials and methods). Tie2-Cre-mediated removal of the floxed NeoSTOP sequences resulted in a mild 1.2fold endothelial Dll4 overexpression that permitted survival of transgenic embryos bearing a single copy of the Dll4 GOF allele (not shown). At E14.5, forced Dll4 expression lead to marginally increased, sub-epicardial vessel (Emcn-positive) coverage ( Figure 4A,F-Source data 1, sheet 10) and vascular malformations similar to those found in Dll4 flox ;Pdgfb-iCre ERT2 and Dll4 flox ;Cdh5-Cre ERT2 mutants ( Figure 1-figure supplement 3E). However, by E16.5 IsoB4-positive (arteries) intramyocardial vessels were substantially decreased ( Figure 4B,F-Source data 1, sheet 10). Accordingly, Dll4 and Efnb2 expression was restricted to smaller caliber vessels in Dll4 GOF transgenics ( Figure 4C). In contrast, superficial Emcn-positive vessels (veins) were more numerous and dense ( Figure 4B,F-Source data 1, sheet 10). N1ICD expression was increased and extended to the prospective veins ( Figure 4D,G-Source data 1, sheet 10), consistent with Notch1 gain-of-function in endothelium. Arterial smooth muscle cell coverage was also substantially reduced ( Figure 4E,G-Source data 1, sheet 10), suggesting defective pericyte and/or smooth muscle differentiation. These coronary vascular defects were associated with below-normal cardiomyocyte proliferation and reduced myocardial thickness (not shown). Therefore, endothelial Dll4 overexpression blocks coronary artery formation, vessel remodeling and maturation, and impairs cardiac growth. These results were surprising, as our loss-of-function data support a pro-arteriogenic role for Dll4.
To support our Dll4 GOF results, we used a transgenic line conditionally overexpressing Mfng

EphrinB2 is required for coronary arteriogenesis and vessel branching
Efnb2 is necessary for arterial-venous differentiation and vascular maturation (Kania and Klein, 2016) and is a Notch target during ventricular chamber development . We find Efnb2 expression in ventricular endocardium at E10.5 (Figure 5-figure supplement 1A) but not in SV endothelium (not shown). At E13.5, Efnb2 was expressed in emerging arteries and prospective veins ( Figure 5-figure supplement 1A), and at E16.5 was confined to arteries (Figure 5figure supplement 1A). Therefore, Efnb2 is expressed dynamically in a pattern similar to Dll4 and Mfng.
Accordingly, the venous marker EphB4 was expressed ectopically in a subset of intramyocardial vessels at E16.5 ( Thus, endocardial EphrinB2 is required for coronary arterial remodeling, vessel structural integrity, and cardiac growth.
Based on the congruence of endocardial/endothelial Jag1, Dll4 and Efnb2 loss-of-function phenotypes, we tested whether EphrinB2 mediates Notch function in coronary angiogenesis. Thus, delivery of a lentivirus expressing Efnb2 to Jag1 flox ;Nfatc1-Cre ventricular explants normalized the reduced endothelial branch number to the number seen in control explants ( Figure 6B,G,K-Source data 1, sheet 14). Likewise, Efnb2 expression restored the elevated number of endothelial branches in Dll4 flox ;Nfatc1-Cre explants to control levels ( Figure 6B,H,K-Source data 1, sheet 14), showing that the branching defect could be normalized in both mutants.  The ventricular explant assay assesses the collective behaviors of endocardium and coronary endothelial outgrowth. In order to determine endothelial behavior exclusively, we performed capillary tube formation assays with primary human endothelial cells (HUVEC). shRNA directed against JAG1, DLL4 or EFNB2 were lentivirally-transduced into HUVEC. This resulted in reduction of mRNA levels of 30% and 60% for JAG1 and DLL4, respectively ( Figure  We next tested whether EPHRINB2 acts downstream of JAG1 and DLL4 in capillary tube formation. JAG1 knockdown inhibited capillary tube formation and decreased network complexity as determined by measuring EC junctions, nodes, segments and pieces that were below control, while the 'total isolated branches length' was above control ( Figure 7B,C-Source data 1, sheet 15). However, these parameters were restored to control levels in presence of the EFNB2 transgene ( Figure 7B,C-Source data 1, sheet 15), suggesting that EPHRINB2 compensates for JAG1 knockdown in this assay. In contrast, knockdown of DLL4 increased capillary tube formation and network complexity (junctions, nodes, segments and pieces; Figure 7B,C-Source data 1, sheet 15), while measurements were restored and even went beyond control by overexpressing EFNB2 (Figure 7B, C-Source data 1, sheet 15). Thus, EFNB2 overexpression not only compensates for the absence of DLL4 in this assay, but has an added effect as well.
Taken together, our observations with ventricular explants and HUVEC indicate that JAG1 and EPHRINB2 promote coronary vessel sprouting and branching, whereas DLL4, MFNG and NOTCH1 inhibit these processes. Moreover, EPHRINB2 mediates signaling from both JAG1 and DLL4 to regulate endothelial branching during coronary angiogenesis.

Discussion
The origin of the coronary endothelium from a venous source has prompted the suggestion that arteries need to be 'reprogrammed' from veins (Red-Horse et al., 2010). We found Jag1, Dll4, Mfng, and N1ICD expression in SV endothelium and sub-epicardial capillaries, with Dll4 and Mfng expression maintained subsequently in prospective veins. These Notch pathway elements are coexpressed with the venous marker endomucin, which is also expressed in the endocardium and SV (Cavallero et al., 2015). This implies that nascent subepicardial vessels have a mixed arterial/venous identity. During embryonic angiogenesis, Notch is necessary for artery-vein specification-the reversible commitment of ECs to arterial or venous fate-before the onset of blood flow (Swift and Weinstein, 2009). Our finding of SV endothelial cell heterogeneity in relation to arterial-venous identity is consistent with the notion that early vascular beds are phenotypically plastic during embryonic Figure 5 continued expressing coronaries. Scale bars, 100 mm. (D) Immunohistochemistry for EphB4 (red) and IsoB4 (green) on E16.5 control and Efnb2 flox ;Nfatc1-Cre heart sections. Dapi-counterstain (blue). Yellow arrowheads indicate EphB4-stained vessels. Scale bar, 50 mm. (E) Immunohistochemistry for SMA (red) and Notch3 (green) on E16.5 control and Efnb2 flox ;Nfatc1-Cre heart sections. Dapi-counterstain (blue). Arrow points to a coronary artery stained by SMA and Notch3. Microscope: Nikon A1-R. Software: NIS Elements AR 4.30.02. Build 1053 LO, 64 bits. Objectives: Plan Apo VC 20x/0,75 DIC N2 dry; Plan Fluor 40x/1,3 Oil DIC H N2 Oil. (F) Quantified data for control and E16.5 WT and Efnb2 flox ;Nfatc1-Cre hearts: E15.5, area covered by coronary arteries (IsoB4positive vessels), area covered by veins (Emcn-positive vessels). Data are mean ± s.d, (n = 3 control embryos and n = 3 mutant embryos), and percentage of N1ICD-stained nuclei in endothelium as percentage (%) of total nuclei Data are mean ± s.d. (n = 3 sections from 3 control embryos and n = 3 sections from 4 mutant embryos); E16.5, SMA-Notch3 co-immunostaining in coronary arteries. Data are mean ± s.d. (n = 3 sections from three control embryos and n = 3 sections from 3 mutant embryos). *p<0.05, **p<0.01 by Student's t-test; n.s., not significant. The online version of this article includes the following figure supplement(s) for figure 5:   Figure 6 continued on next page development (Moyon et al., 2001;Chong et al., 2011;Fish and Wythe, 2015). Our results are consistent with recently published data showing that pre-arterial cells expressing Notch pathway elements are present in coronary endothelium (Su et al., 2018), prior to blood flow onset.
We show that Notch ligands Jag1 and Dll4 are required for SV sprouting angiogenesis, further supporting the notion that coronary arteries are specified by Notch prior to blood flow (Su et al., 2018) ( Figure 8A). Dll4 is the key Notch ligand activator regulating embryonic (Duarte, 2004;Gale et al., 2004;Krebs et al., 2004;Benedito et al., 2008) and postnatal retinal angiogenesis (Hellströ m et al., 2007;Lobov et al., 2007), whereas Jag1 antagonizes Dll4-Notch1 signaling (Benedito et al., 2009) and acts downstream of Dll4-Notch1 to promote smooth muscle differentiation (Pedrosa et al., 2015). Jag1 or Dll4 inactivation in the SV results in arrested and excessive angiogenesis, respectively, resembling the situation in the retina (Benedito et al., 2009) ( Figure 8B). Moreover capillary 'entanglements' suggestive of vascular malformations were present in the early plexus of Jag1 flox ;Nfatc1-Cre and Dll4 flox ;Nfatc1-Cre mutants. Vascular malformations in Notch mutants have been attributed to unresolved intermingling of arteries and veins during differentiation, or to failed maintenance of arterial and venous identities in the vascular bed (Gale et al., 2004;Krebs et al., 2004). The presence of these malformations in Jag1 flox ;Nfatc1-Cre and Dll4 flox ; Nfatc1-Cre mutants could therefore indicate that endothelial progenitors begin to differentiate into arteries and veins very soon after exiting the SV.
Consistent with the known roles of the Notch ligands in angiogenesis, the hierarchical organization of the coronary vascular tree was profoundly altered in the late-induced Jag1 and Dll4 mutants, implying that both Notch ligands are required for high-order complexity of the coronary vessels ( Figure 8C). Jag1 or Dll4 inactivation during coronary arterial remodeling results in smaller diameter arteries, consistent with Notch promoting vascular remodeling and arterial fate commitment, as described in other developmental settings ( Figure 8D; (Duarte, 2004;Gale et al., 2004;Krebs et al., 2004). However, gain of Notch function also led to smaller caliber arteries ( Figure 8D), when the opposite outcome was anticipated (Uyttendaele et al., 2001;Trindade et al., 2008;Krebs et al., 2010). This result implies that non-physiological variations of Notch activity lead to the arrest of coronary artery differentiation. Dll4 mutants also showed coronary vessel hemorrhaging reminiscent of that found in vascular disorders (Park-Windhol and D'Amore, 2016) or tumors (Goel et al., 2011) and indicative of disrupted endothelial integrity.
We reasoned that the coronary maturation defects observed in Jag1 or Dll4 mutants are due, at least in part, to the loss of EphrinB2 function given that Efnb2 is a direct Dll4-Notch signaling target in ECs (Iso et al., 2006) and endocardium during ventricular development . Moreover, EphrinB2 and VEGF are functionally linked during angio-and lymphangiogenesis; EphrinB2 is a direct activator of VEGFR2 and VEGFR3, and therefore cooperates in the mechanism leading to tip cell extension and vessel sprouting (Sawamiphak et al., 2010;Wang et al., 2010). Thus, Efnb2 inactivation leads to coronary artery remodeling defects, similar to those resulting from Jag1 or Dll4 inactivation, suggesting that EphrinB2 functions downstream of Notch to promote coronary arterial remodeling ( Figure 8D). This notion is supported by the angiogenic ventricular explant and EC capillary tube assays, in which opposite effects of Jag1 or Dll4 deficiency on vessel branching are rescued by transduction of an EFNB2-expressing lentivirus, identifying EphrinB2 as a Notch effector during coronary artery development.
A common feature among the endocardial or endothelial Notch loss-and gain-of-function models analyzed in our study is the thin ventricular wall ( Figure 8B,D). Of interest is that myocardial inactivation of Jag1, or combined inactivation of Jag1 and Jag2, or Mib1, leads to thinner ventricular walls, accompanied by reduced cardiomyocyte proliferation, disrupted ventricular chamber patterning, and cardiomyopathy (Luxán et al., 2013;D'Amato et al., 2016). In contrast, chamber patterning is maintained in endocardial Jag1, Dll4, and Efnb2 mutants and endothelial Dll4 and Jag1 mutants. Our results suggest that disturbed ventricular wall growth in the earlier E11.5-E12.5 Jag1 flox ;Nfatc1-Cre is caused by altered endocardial-myocardial signaling, as suggested for Dll4 flox ;Nfatc1-Cre    et al., 2016). In the later E14.5-E16.5 Jag1 flox ;Pdgfb-iCre ERT2 , Dll4 flox ;Pdgfb-iCre ERT2 , Jag1 flox ;Cdh5-Cre ERT2 and Dll4 flox ;Cdh5-Cre ERT2 mutant embryos, a thinner ventricular wall would be due to the lack of a well-formed coronary plexus ( Figure 8D). At E12.5-E13-5, myocardial growth may also depend on 'angiocrine signals' from the un-perfused primitive coronary plexus (Rafii et al., 2016), as the diffusion limit of oxygen and nutrients from the endocardium is reached during the transition from endocardial to coronary myocardial perfusion. Therefore, ventricular compaction relies on two interconnected Notch-dependent processes: patterning and maturation of the chamber myocardium, and timely development of a functional coronary vessel network, as previously suggested (D'Amato et al., 2016). This may be clinically relevant to the study and treatment of cardiomyopathies.    (Nowotschin et al., 2013), Tie2-Cre (Kisanuki et al., 2001), Nfatc1-Cre (Wu et al., 2012), Pdgfb-iCre ERT2 (Wang et al., 2010), Cdh5-Cre ERT2 (Wang et al., 2010), Dll4 flox (Koch et al., 2008), Jag1 flox (Mancini et al., 2005), Efnb2 flox (Grunwald et al., 2004), Notch1 flox (Radtke et al., 1999), Mfng GOF (D'Amato et al., 2016). To generate the R26CAGDll4 GOF transgenic line, a full-length mouse Dll4 cDNA was obtained from clone IMAGE 6825525. The sequence was PCR amplified with Phusion High-Fidelity DNA Polymerase (NEB) and primers containing BamHI and ClaI sites and was cloned in-frame with 6 Myc Tag epitopes into the BamHI and ClaI sites of a pCS2-MT plamid. The Dll4-MT fragment was subcloned into pCDNA3.1 with BamHI and EcoRI and excised with BamHI and NotI. We then modified pCCALL2 (Lobe et al., 1999) by cloning new XbaI sites before and after a CAG cassette, which includes a CMV enhancer/b-actin promoter and a rabbit b-globin polyA signal. The KpnI and NotI PCR Dll4-MT was cloned into the BglII-NotI sites of the modified pCCALL2. The XbaI-cassette containing CAG-loxP-b-Geo-loxP-Dll4-IRESeGFP was obtained by digestion and cloned into the XbaI site of the pROSA26-1 plasmid (Soriano, 1999). The final construct was linearized with XhoI and electroporated into R1 mESCs derived from a cross of 129/Sv x 129/Sv-CP mice (Nagy et al., 1993). After G418 (200 mg/ml) selection for 7 days, 231 clones were picked. Homologous recombination was identified by Southern blot of EcoRV-digested DNA and hybridized with 5' and 3' probes. About 25% of the clones were positive, and we selected three clones to confirm karyotype. One positive clone was injected into C57/BL6C blastocysts to generate chimaeras that transmitted the transgene to their offspring. The resulting founders were genotyped by PCR of tail genomic DNA using primers targeting the R26 locus before and after the cloning site and the transgene polyA signal. Center: cross-section through the ventricular wall. SV-derived ECs invade the sub-epicardial space over the myocardium to cover the heart dorsally. Some pre-artery ECs invade the myocardium and begin to differentiate into arteries. Right: during EC sprouting and proliferation, a regulatory balance between Dll4/Mfng and Jag1 modulates Notch signaling output downstream of Vegf. (B) Left: E12.5 Jag1 flox ;Nfatc1-Cre (Jag1LOF; top) exhibit arrested coronary angiogenesis, Dll4 flox ;Nfatc1-Cre mutants (Dll4LOF; bottom) display increased angiogenesis, and both mutants display vascular malformations. The zoomed views show details of a sub-epicardial vessels with increased N1ICD expression in the Jag1LOF mutants (red), and decreased expression in the Dll4LOF mutants (blue). Center: the sub-epicardial capillary plexus is either poorly developed in Jag1LOF mutants or overdeveloped in Dll4LOF mutants, and with vascular malformations. Mutant hearts have a thin myocardial wall, but maintain myocardial patterning (greybrown boundary). Right: Endocardial Jag1 deletion disrupts the regulatory balance with Dll4/Mfng, leading to increased N1ICD and arrested migration and proliferation. In contrast, endocardial Dll4 deletion results in decreased N1ICD signaling, and increased network complexity and proliferation. (C,D) Embryonic stages E14.5-E16.5. Arterial differentiation and plexus remodeling. (C) Left: coronary vasculature at the dorsal aspect of the wild-type heart. Zoomed view showing a coronary artery with an inner layer of ECs expressing N1ICD (red) and an outer layer of smooth muscle cells (yellow). Center: capillary differentiation and patterning give rise to large veins sub-epicardially and large arteries intra-myocardially. Right: systemic blood flow activates Dll4/Jag1/Mfng/N1ICD/EphrinB2 signaling to drive terminal arterial differentiation and remodeling. (D) Left: in coronary endothelial-Dll4 flox ;Pdgfb-iCre ERT2 or -Jag1 flox ;Pdgfb-iCre ERT2 mutants, the coronary vasculature is mis-patterned. Zoomed view detailing near absence of arteries in Dll4 mutants, and comparatively smaller caliber coronary artery in Jag1 mutants. The arteries in both mutants have an inner layer of a 'leaky' endothelium and an outer layer of poorly differentiated perivascular cells (white). Center: Dll4 mutants are characterized by the near-absence of coronary arteries while arteries in Jag1 mutants are decreased. In either case, veins are unaffected. Mutant hearts have a thin myocardial wall but maintain myocardial patterning (grey-brown boundary). Right: Coronary endothelial-Jag1 or Dll4LOF leads to increased or decreased N1ICD respectively, causing arrested arterial differentiation and remodeling. Inactivation of Efnb2 leads to a similar phenotype. Endothelial Notch gain-of-function (GOF), resulting from Dll4 or Mfng overexpression, also leads to increased N1ICD and disrupted arterial differentiation and remodeling.

Histology and in situ hybridization
Hematoxylin and eosin (H & E) staining and in situ hybridization (ISH) on sections were performed as described (Kanzler et al., 1998). Details of probes will be provided on request.
For EdU immunofluorescence, pregnant females at E12.5 were injected intraperitoneally with 100 ml EdU nucleotides (2 mg/ml in PBS). The mice were euthanized 1 hr later and fixed in 4% PFA. EdU incorporation was detected with the Click-iT EdU Imaging Kit (Thermo Fisher Scientific, C10340) according to manufacturer's instructions. For IsoB4 and Endomucin immunostaining on E11-4-E16.5 embryos, imaging by confocal z-projection of the deeper section of the myocardium captures the arteries (IsoB4-stained) with the outer section corresponding to the veins (IsoB4+Endomucinstained). N1ICD embryos were fixed for 5 hr at 4˚C in frozen Methanol, washed and incubated 20 min at 98˚C in DAKO target retrieval solution pH6 (Agilent). After washing in H 2 O, hearts were fixed again in acetone at À20˚C for 10 min. After washing in PBT, the hearts were blocked in 10% FBS, 5% BSA, 0.4% TritonX100 for 3 hr at RT under gentle rocking. Hearts were incubated with primary N1ICD-antibody at 1:500 for 2 days at 4˚C and allowed to settle at RT for 1 hr. This was followed by washing 5 hr in PBS/0.4% TritonX100 at RT. Hearts were then incubated in presence of anti-rabbit HRP (1:500), DAPI (1:1500) and anti-VE cadherin (1:500) overnight at 4˚C. After resting for 1 hr at RT, hearts were washed in PBT, and 30 min in presence TSA (tyramides) diluted 1:200 in PBS. Quantifications were carried out using ImageJ software. The main coronary trees were selected with the tool 'Freehand selection' manually and measured directly from the autoscaled images obtained by Z-projection. The selected area was quantified in mm 2 . Confocal microscopy analysis was carried out on a Nikon A1R or Leica LAS-AF 2.7.3.
Images were processed using ImageJ software. For p21 and N1ICD quantification, the number of positive nuclei was divided by the total number of nuclei counted on sections (!3). For aSMA and Notch3 quantification, the number of mural cells surrounding intramyocardial vessels expressing both SMA+ and Notch3+ (double positives) was counted and divided by the total number of intramyocardial vessels defined by Iso B4 staining in the same section. For SM22a and Notch3 quantification, the number of double-SM22 and Notch3 positive cells was divided by the total number Notch3-positive cells counted on sections (!3).

Mouse ventricular explant culture and immunofluorescence
Heart ventricular explants were performed as previously described with minor modifications (Wu et al., 2012). Ventricles were dissected from E10.5 or E11.5 embryos (with removal of the outflow tract and atria), rinsed with PBS to remove circulating cells, and placed in Nunc four-well plates. Matrigel (Corning Matrigel Basement Membrane Matrix, *LDEV-Free, 10 mL growth factor reduced, BD Biosciences 354234) was diluted 1:1 with DMEM plus 10% FBS and 10 ng/ml Vegf (Human-Vegf 165 Peprotech). Each well had a total volume of 400 ml, and 3-4 hearts were cultured for 6 days and then fixed in 4% PFA4% 10 min and washed twice with PBS1X for 15 min. Explants were permeabilized with 0.5% Triton X-100 for 1 hr, blocked with Histoblock solution (FBS Histoblock) for 2-3 hr at RT, and incubated with anti-CD31 1:100 (Purified Rat Anti-Mouse CD31 BD Biosciences Pharmingen) overnight at 4˚C. Anti-rat biotinylated was used as secondary antibody and diluted in BSA 1:150. Staining was amplified with the ABC kit and 3 min of TSA (tyramides) diluted 1:100 in PBS.

Quantification of compact myocardium thickness
The method used was a modification of that described by Chen et al. (2009);Yang et al. (2012). Briefly, 7 mm paraffin sections from E12.5, E15.5 and E16.5 wild type (WT) and mutant hearts were stained with anti-CD31 or Endomucin (Emcn) and anti-MF20 antibodies to visualize ventricular structures. Confocal images were obtained with a NIKON A1R confocal microscope. Measurements were made using ImageJ software. In E15.5 and E16.5 heart sections, endocardial cells were stained with anti-Emcn and myocardium with anti-cTnT. Left and right ventricles were analyzed separately. For each measurement, settings were kept constant for all images, using the scale bar recorded in each image as the reference distance. The thickness of the compact myocardium was measured by dividing the ventricle into left and right parts. Several measurements were taken in each region and the mean was expressed in mm.

Quantification of Jag1 and Dll4 expression in sections
Quantification of ISH signal were carried out using ImageJ software. Compact myocardium area was selected manually with the tool 'Polygon selection'. Coronary area was identified in this selection by color intensity using 'Color Threshold plugin' with the settings: Hue: Min = 145 Max = 199, Saturation: Min = 60 Max = 255, Brightness: Min = 0 Max = 255. Positive coronary area for Dll4 or Jag1 expression was calculated dividing the coronary area positive for each gene by the total compact myocardium area. Statistical analysis was assessed by Student's t-test.

India ink perfusion
Embryos were collected from E15.5 to E18.5 in PBS containing 2 mg/ml heparin. The thoracic cavities were immediately cut and transferred to DMEM containing 10%FBS and heparin on ice. Hearts were carefully dissected from the surrounding tissue and then placed on an inverted Petri dish and gently dried to avoid movement during perfusion. India ink or red tempera was diluted in PSB/heparin and injected from the ascending aorta with a borosilicate glass tube thinned to the appropriate diameter and attached to a mouth pipette. India ink was slowly injected by minute puffs of breath during diastolic intervals. Hearts were fixed in 4% PFA, dehydrated, cleared in BABB (benzyl alcohol: benzyl benzoate, 1:1) and imaged with a stereomicroscope.

Culture and infection on HUVEC
Human umbilical vein endothelial cells (HUVECs) were maintained in supplemented EGM2 medium (EGM2 Bulletkit, K3CC-3162 Lonza). Cells were transduced on suspension at a multiplicity of infection (m.o.i.) of 80 with combinations of 2 shRNAs against JAG1 or DLL4 together with GFP or EFNB2 overexpressing lentivirus and seeded onto 24-well plates or 12-well plates at a density of 3 Â 104 or 6 Â 104 cells/well respectively. Infection efficiency was near 100%. HUVECs were harvested after 48 hr. Gene expression analysis was performed using KiCqStart SYBR Green predesigned primers (Sigma-Aldrich).

Angiogenesis assays on matrigel
Angiogenesis in vitro was investigated using endothelial cell tube formation assay on Matrigel (#354234 Corning) as previously described (Grant et al., 1991). Briefly 50 ml of ice cold Matrigel was coated on a 96 micro-well plate as a base for tube formation. After allowing the gel to settle in the incubator for 30 min at 37˚C, 5% CO2, HUVECs in basic EGM medium (EBM2 H3CC-3156, Lonza) were seeded in triplicate at a density of 12.5 Â 104 on Matrigel and incubated at 37˚C, 5% CO2. After 5 hr, tube formation was monitored by phase-contrast microscope. The tube networks were quantified using NIH Image J with Angiogenesis plugin software (http://image.bio.methods.free.fr/ ImageJ/?Angiogenesis-Analyzer-for-ImageJ). Excel data were represented on Graphpad Prism seven as mean with SD. For statistical analysis a one-way ANOVA and Tukey post-hoc tests were performed.

Statistical analysis
Statistical analysis was carried out using Prism 7 (GraphPad). All statistical tests were performed using two-sided, unpaired Student's t-tests, except for Figures 1, 2 and 7 and figures Supplements 4, 9 and 12 where we performed one-way ANOVA to assess statistical significance with a 95% confidence interval, where numerical data are presented as mean ± SD; results are marked with one asterisk (*) if p<0.05, two (**) if p<0.01, and three (***) if p<0.001. Sample size was chosen empirically according to previous experience in the calculation of experimental variability. No statistical method was used to predetermine sample size. All experiments were carried out with at least three biological replicates. The numbers of animals used are described in the corresponding figure legends. Animals were genotyped before the experiment and were caged together and treated in the same way. Variance was comparable between groups throughout the manuscript. We chose the appropriate tests according to the data distributions. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

RNA-sequencing
RNA was isolated at E12.5, from whole hearts of Jag1 flox ;Nfatc1-Cre and Dll4 flox ;Nfatc1-Cre embryos, as well as their WT counterparts (four replicate samples for each condition, pooling four embryos per sample, in all cases). At E15.5, RNA was isolated from ventricles of Dll4 flox ;Cdh5-Cre ERT2 (three replicate samples, pooling three embryos per sample), as well as their WT counterparts (four replicate samples, pooling three embryos per sample). RNA-Seq data for Jag1 flox ;Nfatc1-Cre, Dll4 flox ;Nfatc1-Cre, and Dll4 flox ;Cdh5-Cre ERT2 experiments was generated by CNIC's Genomics Unit. RNA-Seq sequencing reads were pre-processed by means of a pipeline that used FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) to assess read quality and Cutadapt v1.6 (Martin, 2011) to trim sequencing reads, thus eliminating Illumina adaptor remains, and to discard reads shorter than 30 bp. Resulting reads were mapped against the mouse transcriptome (GRCm38 assembly, Ensembl release 76) and quantified using RSEM v1.2.20 (Li and Dewey, 2011). Around 80-90% of the reads participated in at least one reported alignment. Expected expression counts calculated with RSEM were then processed with an analysis pipeline that used Bioconductor package Limma (Ritchie et al., 2015) for normalization and differential expression testing in six pairwise contrasts involving mutant versus WT comparisons. We discarded two of the Dll4 flox ;Nfatc1-Cre samples based on preliminary analyses of Dll4 expression levels and clustering patterns in diagnostic principal component analysis plots. Changes in gene expression were considered significant if associated with a Benjamini-Hochberg (BH) adjusted p-value<0.05. The number of differentially expressed genes detected in comparisons between Jag1 flox ;Nfatc1-Cre, Dll4 flox ;Nfatc1-Cre and Dll4 flox ;Cdh5-Cre ERT2 and their control counterparts was 205, 257 and 156, respectively. Enrichment analyzes were performed with IPA (Ingenuity Pathway Analysis, Qiagen, http://www.ingenuity.com). IPA was used to identify collections of genes associated with canonical pathways, common upstream regulators, or functional terms significantly overrepresented in the sets of differentially expressed genes; statistical significance was defined by Benjamini-Hochberg adjusted p-value<0.05. Circular plots summarizing the association between genes and enriched pathways, upstream regulators, and functional terms were generated with GOplot (Walter et al., 2015).
of differentially expressed genes that are targets of the regulator. Downstream Effect analyses results, in sheets labelled as 'CardSystDevFunc', are restricted to functions of category 'Cardiovascular System Development and Function'. For each functional term, tables describe enrichment p value, predicted activation state, activation z-score and the collection of differentially expressed genes associated to that particular function. Positive and negative z-score values suggest activation or inhibition of the corresponding upstream regulator or function in the mutant or control condition, respectively; abs(z-score)>2 and p value < 0.05 are considered significant.
. Supplementary file 4. List of primary and secondary antibodies used in this study to immunodetect proteins in whole-mount or paraffin sections.
. Transparent reporting form

Data availability
Sequencing data have been deposited in GEO under accession code GSE110614.
The following dataset was generated: