Notch-independent RBPJ controls angiogenesis in the adult heart

Increasing angiogenesis has long been considered a therapeutic target for improving heart function after injury such as acute myocardial infarction. However, gene, protein and cell therapies to increase microvascularization have not been successful, most likely because the studies failed to achieve regulated and concerted expression of pro-angiogenic and angiostatic factors needed to produce functional microvasculature. Here, we report that the transcription factor RBPJ is a homoeostatic repressor of multiple pro-angiogenic and angiostatic factor genes in cardiomyocytes. RBPJ controls angiogenic factor gene expression independently of Notch by antagonizing the activity of hypoxia-inducible factors (HIFs). In contrast to previous strategies, the cardiomyocyte-specific deletion of Rbpj increased microvascularization of the heart without adversely affecting cardiac structure or function even into old age. Furthermore, the loss of RBPJ in cardiomyocytes increased hypoxia tolerance, improved heart function and decreased pathological remodelling after myocardial infarction, suggesting that inhibiting RBPJ might be therapeutic for ischaemic injury.

A ngiogenesis in the adult heart is tightly regulated in order to maintain cardiac function in response to workload 1 . The mechanisms that coordinate angiogenesis and workload are poorly understood, but involve the regulated secretion of angiogenic factors from cardiomyocytes 2,3 .
Increasing angiogenesis is of therapeutic interest to treat chronic myocardial ischaemia, 4,5 as well as acutely after myocardial infarction since many patients have inadequate perfusion despite restoration of coronary artery blood flow 6 . Despite pre-clinical proof of concept, clinical trials of single factor, gene and cell-based therapies have provided inconsistent results, and identifying a definitive means of inducing clinically useful therapeutic angiogenesis remains elusive [4][5][6] . A challenge in achieving therapeutic angiogenesis is that the formation of functional vessels involves the coordinated activities of numerous factors, some of which promote endothelial cell proliferation and vessel sprouting (for example, vascular endothelial growth factor (VEGF) and fibroblast growth factor (FGF)), while others drive vessel remodelling and maturation (for example, angiopoietins) 7 .
Since pressure overload increases myocardial angiogenesis and activates Notch signalling 8,9 , we hypothesized that Notch and its transcriptional effector RBPJ (recombination signal binding protein for immunoglobulin kappa J region) might function in cardiomyocytes as a high level modulator of myocardial angiogenesis. RBPJ is ubiquitously expressed, and can either activate genes by forming a complex with the Notch intracellular domain (NICD) when Notch is active 10 , or silence an overlapping but non-identical set of genes by recruiting co-repressors in the absence of Notch signalling [11][12][13] .
Our studies revealed that RBPJ represses the production of pro-angiogenic and angiostatic factors in cardiomyocytes of the unstressed heart, such that the adult cardiomyocyte (ACM)specific deletion of rbpj increases microvessel density, confers physiological tolerance to hypoxia and protects the heart from ischaemic injury. Contrary to expectations, activation of Notch itself did not induce an angiogenic programme. We present evidence that RBPJ controls angiogenic factor gene expression by direct binding and inhibition of HIF1a and 2a. Thus, RBPJ has a novel Notch-independent role as a repressor of hypoxia responsiveness and myocardial angiogenesis.

Results
RBPJ acts as a transcriptional repressor in healthy ACM. We first investigated the location and activity of RBPJ in the mouse adult heart. Nuclear-localized and faint cytoplasmic RBPJ is readily detected in the adult heart, both in cardiomyocytes and non-myocytes, and its level does not change substantially following either chronic pressure overload by thoracic aortic constriction (TAC) or myocardial infarction (Fig. 1a,b; Supplementary Fig. 1). In contrast, RBPJ was nearly undetectable basally by immunostaining with the monoclonal antibody T6709 (ref. 14) (Fig. 1d), but was strikingly evident in response to a doxycyclin-inducible, transgenic NICD (iNICD) 15 (Fig. 1c,f) and after TAC or myocardial infarction (MI) ( Fig. 1e; Supplementary  Fig. 1c-f) coinciding with elevated expression of Notch ligands ( Supplementary Fig. 2a-d). T6709-detection correlates with Notch activation 14 , suggesting that RBPJ in the heart might repress gene expression basally and mediate Notch signalling after MI and pressure overload.
To investigate the potential function of RBPJ in the heart, we deleted Rbpj in ACM with Myl2-Cre (cKO). cKO mice had normal viability consistent with post-natal deletion (Supplementary Table 1; Supplementary Fig. 3a,b). Deletion increased expression of the endogenous Notch/RBPJ target genes Hes1 and Hey2 identically to the effect of inducing Notch activation (iNICD) in unstressed hearts (Fig. 1g,h), and increased the content of activating tri-methylated (K4) histone-H3 (H3K4me 3 ) in a proximal region of the endogenous Hes1 and Hey2 promoter containing RBPJ-binding motifs (Fig. 1i). In vitro, RBPJ overexpression repressed a Hes1 promoter-luciferase reporter (Fig. 1j). Finally, Hes1 messenger RNA (mRNA) content was higher in isolated cKO relative to wild-type (WT) cardiomyocytes, and could be induced by adenoviral delivery of NICD in an RBPJ-dependent manner (Fig. 1k). These data indicate that RBPJ acts as a transcription repressor basally in cardiomyocytes, but is converted to an activator upon pressure overload or MI (Fig. 1l).
Lack of RBPJ induces angiogenesis in the heart. Next, we evaluated the consequences of rbpj deletion. cKO hearts showed no evidence of cardiomyocyte pathology (see Supplementary  Fig. 4 for a detailed phenotypic characterization), and responded normally to pressure overload for all parameters analyzed, including cardiac function (Supplementary Table 2) and hypertrophy ( Supplementary Fig 5). However, the loss of RBPJ significantly increased the number of capillaries relative to cardiomyocytes (CC ratio) from 1.6±0.27 in WT hearts to 2.1 ± 0.28 in cKO hearts ( Fig. 2a-c). cKO capillaries were morphologically normal even into old age (20 months, Fig. 2d,e), and were perfused by ILB4 lectin to the same degree as WT capillaries (Fig. 2f,g). Fibrinogen was not detected outside the vessel walls ( Fig. 2h-k), indicating that the capillaries were not leaky, in contrast to capillaries following myocardial infarction as a control (Fig. 2j,k), or following chronic hypoxia-inducible factor (HIF) stabilization 16,17 . Thus, deletion of rbpj in cardiomyocytes increased the density of functional capillaries without inducing cardiac pathology.
Capillary density normally increases in response to chronically elevated blood pressure. In WT mice, 2 weeks of TAC significantly increased the CC ratio to 2.2 ± 0.33. Myocardial microvascularization did not increase after TAC in the cKO-RBPJ mice beyond the already elevated basal level (1.9±0.17 CC ratio, statistically indistinguishable from the basal level). One possible explanation is that TAC signals through Notch/RBPJ to promote angiogenesis. Contrary to our initial hypothesis, doxycyclineactivated iNICD did not affect the CC ratio (Fig. 2a), suggesting that TAC-induced microvascularization does not involve Notch.
To probe the mechanism responsible for the increased microvascularization, we profiled the expression of mRNAs encoding 29 secreted angiogenic factors. Five were upregulated in the cKO relative to WT, including factors involved in sprouting and endothelial cell migration (for example, VEGF-A and PGF), or stabilization (THBS2), and just a single factor (TGFb3) was downregulated (Fig. 2l). iNICD did not upregulate any angiogenic factor mRNAs (but downregulated Vegfa, Tnf and Ang2) despite upregulating the canonical Notch/RBPJ target genes Hes1 and Hey2 (Fig. 1j). Although consistent with no effect of iNicd on the CC ratio, the result was puzzling since the angiogenic factor genes behaved differently than canonical Notch/RBPJ target genes Hes1 and Hey2.
We investigated the divergent effects of the cKO and iNICD genotypes on angiogenic factor expression using Vegfa as a paradigm. Consistent with a Notch-independent role, VEGF-A protein levels were elevated in cKO, but not iNICD, heart tissue relative to WT controls (Fig. 2m,n; Supplementary Fig. 6). Culture media conditioned by cKO isolated ACM showed an increased ability to form microvascular endothelial tubes in vitro relative to WT-conditioned media, and this increased activity was blocked by the VEGF receptor inhibitor (carbozantinib) (Fig. 2o). Taken together, the in vitro and in vivo data support the model that RBPJ inhibits VEGF-A dependent angiogenesis. Chromatin immunoprecipitation revealed increased content of histone-H3 modifications (K9/K18Ac 2 and K4me 3 ) in the endogenous Vegfa promoter in a region B1.5 kb upstream of the start site of transcription that contains both RBPJ and HIF consensus motifs ( Fig. 2p; Supplementary Table 3). These epigenetic marks indicate gene activation in the cKO cardiomyocytes. Activation was reversed by the overexpression of RBPJ, and co-transfection with a truncated-form of SHARP that physically sequesters RBPJ 18 completely eliminated repression (Fig. 2q). Thus, RBPJ represses Vegfa gene expression under unstressed conditions, and deleting RBPJ leads to Vegfa gene activation (Fig. 2r).
Cardioprotection and hypoxia tolerance in rbpj cKO hearts. Given the increased CC ratio in the cKO heart, we evaluated the hemodynamic consequences deleting Rbpj. In addition to the cKO genotype, we generated a tamoxifen-inducible Myh6-Cre 19 knockout (icKO; Myh6 mER-Cre-mER/ þ , Rbpj fl/fl ) to afford precise control over the timing of Rbpj deletion. Blood flow within left ventricular myocardium was greater in the icKO and cKO hearts than in WT controls (Fig. 3a). icKO and cKO mice (along with genotype-specific WT controls) were subjected to acute hypoxia by gradually decreasing ambient O 2 from 21 to 5%. The different groups (cKO, icKO and their WT controls) had similar hemodynamic values under normoxic conditions (21% O 2 ) (Supplementary Table 4). At 5% O 2 , vascular resistance (VR) increased as expected, but was significantly lower in cKO and icKO than WT mice (Fig. 3b), correlating with elevated mean arterial pressure (MAP) and cardiac output (CO) (Fig. 3c-f).
Although not statistically significant, O 2 delivery at baseline and 5% O 2 was greater in the cKO and icKO relative to WT (Fig. 3g). Myocardial hypoxia (measured on histological section) in the cKO and icKO was consistently diminished by 10-20% relative to WT controls (Fig. 3h,i). Thus, loss of RBPJ in ACM correlated with increased myocardial oxygenation and CO under hypoxia.
Since stimulating angiogenesis has been considered a potential therapy for ischaemic disease 5,6,20 , we evaluated the response of icKO mice to MI by serial echocardiography and histology (Fig. 3j). The icKO hearts had increased CC ratio at baseline and preserved cardiac function (Fig. 3l) and morphology (Fig. 3m) relative to WT hearts post-MI (Fig. 3l,  interventricular septum (IVS) since the left ventricle (LV) free wall was injured by the MI. The extent of fibrotic scar formation was also diminished significantly, from 25±4.3% of the total LV area in WT to B8 ± 3.0% in icKO hearts (Fig. 3n). The area of the myocardium that was hypo-perfused acutely (24 h) post-MI was identical in both genotypes, quantified by FITC-ILB4 perfusion (B25% of total LV area) (Fig. 3o), and there was no effect on arterial arborization ( Supplementary Fig. 4  Notch target genes in ventricles from cKO, WT and doxycycline-induced iNICD mice by qRT-PCR normalized to ActB. Volcano plots portray the cKO to WT ratio or iNICD to WT ratio relative to P-value. Red and green points indicate statistically significant (Pr0.05, n ¼ 4) induction or repression of gene expression, respectively; blue points indicate no statistically significant difference. (m,n) VEGFA 165 protein quantification by ELISA in WT, cKO and iNICD heart tissue (m) and conditioned media from WT and cKO-cultured ACM n ¼ 4 (n). (o) Conditioned media from WT and cKO ACM were applied to HMEC1 cells and network formation quantified incubation in the absence or presence of the VEGF receptor inhibitor carbozantinib (2 mM). n ¼ 3. (p) Histone marks in the endogenous Vegfa locus. ChIP with control IgG (white bars) and anti-H3K9/18Ac or anti-H3K4me 3 (black bars) from isolated ACM was amplified using primers spanning RBPJ and HIF consensus binding motifs (schematic) and results normalized to input DNA. Apical sections (towards 6) had significantly smaller hypoxic regions in the cKO (left graph) and icKO (lower graph) hearts compared to their WT controls (j). Box represents mean and 25th and 75th percentiles; whiskers 5th and 95th percentiles. n ¼ 4 mice (j) Experimental protocol for MI. 4-OH tamoxifen was injected to induce recombination of the icKO at daily intervals during a 14-day window ending 5 days prior to LAD ligation (MI), followed by serial echocardiography and histological analysis at endpoint. (k) Microvessel density quantified 6 months after MI on histological sections. Capillary density was examined in the IVS, rather than the anterior wall, since the IVS was not damaged by the infarction. n ¼ 3, 4, 6, 6 for WT, icKO, WT þ MI, icKO þ MI, respectively. (l) LV fractional shortening (%FS) (n ¼ 9 for each genotype and timepoint). P value between genotypes listed. *Po0.05 respect to baseline same genotype. (m,n) Cross-sectional views (m) of WT and icKO hearts 6 months after MI and stained with Masson's trichrome at the level of the papillary muscles. Quantification of the scar area (n) in infarcted myocardium (blue histological stain) expressed as a fraction of LV circumference from 3 parallel cross-sections spaced 700 mm apart (n ¼ 6 WT and 7 icKO; see Methods section). demonstrating that the immediate effect of left anterior descending (LAD) occlusion was the same in both genotypes. We conclude that deleting RBPJ improved tolerance to MI.
RBPJ physically inhibits HIFs. The preceding showed that deleting RBPJ increases angiogenic factor gene expression and microvessel density, correlating with improved cardiac function after MI. We then investigated the mechanism responsible for the puzzling finding that RBPJ represses angiogenic factor genes basally, but that Notch/RBPJ does not induce them. Since hypoxia is a major regulator of angiogenesis 2,6,21 , we examined whether hypoxia responsiveness is altered by the cKO. Exposing isolated ACM to severe hypoxia (1% O 2 ) upregulated 13 and downregulated 4 of the 29 secreted angiogenic factor genes profiled relative to normoxia (21% O 2 ) ( Supplementary Fig. 7). The response at 1% O 2 was nearly identical in WT and cKO genotypes. However, plotting the ratio of gene expression change in cKO relative to WT revealed that the loss of RBPJ sensitized angiogenic genes to induction by moderate hypoxia (6-3% O 2 ) (Fig. 4a). Given that the À 1.5 kb region of the Vegfa promoter contains both RBPJ and HIF consensus binding sites (Fig. 4b, inset; Supplementary Table 3), we hypothesized that RBPJ might interact with HIF to mediate repression. In a luciferase assay, RBPJ repressed induction of the Vegfa promoter by hypoxia (Fig. 4b) as well as by HIF1a or Hif2a (Fig. 4c) in a dosedependent manner.
Repression occurred identically with a mutant RBPJ containing four mutations that block binding to recognition sites in DNA 22 (Fig. 4d), suggesting that RBPJ does not need to interact with DNA in order to inhibit HIF.
Overexpressed GFP-tagged RBPJ in HEK293T cells under hypoxia was co-immunoprecipitated with endogenous HIF1a and HIF2a (Fig. 4e), suggesting a physical interaction between the proteins. Furthermore, a proximity ligation assay (PLA, Fig. 4f) revealed that RBPJ co-localizes with both HIF1a and HIF2a in ACM under normal and hypoxic conditions in vitro. Cytoplasmic and nuclear co-localization was apparent, consistent with the distribution of HIF and RBPJ in both compartments ( Supplementary Fig. 8) and the ability of the RBPJ DNA-binding mutant to antagonize hypoxia and HIF-dependent Vegfa transcription. To control for non-specific interactions, no fluorescent signal was detected in cKO cardiomyocytes (Fig. 4f). Both HIF1a and HIF2a are detectable in WT and cKO mouse hearts (Fig. 4g), although HIF2a is more prominent in nuclei (65% (HIF2a) versus 15% (HIF1a) in WT cells) and increases further in the cKO (80% (HIF2a) versus 15% (HIF1a) in cKO cells) (Fig. 4h; Supplementary Fig. 9). These data indicate that RBPJ physically interacts with HIF proteins basally, and that repression is overcome by HIF stabilization under hypoxia.

Discussion
In summary, RBPJ regulates a paracrine circuit for myocardial angiogenesis (Fig. 4i). RBPJ exerts control over microvessel density by acting within cardiomyocytes to antagonize HIF1a and HIF2a; therefore, desensitizing the angiogenic response to hypoxia. Deleting RBPJ, therefore, increases sensitivity to hypoxia. Attenuating RBPJ in cardiomyocytes is remarkable, in that it benignly increases myocardial vascularization, increases cardiac perfusion and output under hypoxia and preserves cardiac function following MI. There are relatively few instances of RBPJ having a repressive role in mammals [23][24][25] , and most studies have focused primarily on its function as the effector of Notch signalling, which is itself cardioprotective following ischaemic injury 8,9,26 . The effect of RBPJ deletion resembles chronic adaptation of the heart to hypoxia 27 suggesting that attenuating the suppressive function of RBPJ might be therapeutic in the setting of ischaemic injury or chronic overload.

Methods
All animal handling and care followed the NIH Guide for Care and Use of Laboratory Animals. The experimental protocols were approved by Institutional Animal Care and Use Committees of either the Sanford-Burnham-Prebys Medical Discovery Institute, the University of California San Diego or Washington University. Unless otherwise noted, approximately equal numbers of 8-16-weekold male and female mice were used in all studies.
Tamoxifen induction of recombination. 4-OH tamoxifen (Sigma) prepared as 5 mg ml À 1 in corn oil was injected (0.1 ml per mouse) intraperitoneally at daily intervals during a 14-day period that ended 5 days before LAD ligation, as described 29 , or 30 days before physiology tests.
Genetic activation of Notch receptor in adult mouse hearts. aMHCrtTA; tetO-NICD mouse line was previously generated 15 to activate Notch signalling by inducing NICD specifically in ACM. NICD was induced with doxycycline chow (BioServ 200 mg kg À 1 ) during 2 weeks. aMHCrtTA and tetO-NICD littermates fed doxycycline were used for comparison in all conditional gene expression experiments unless otherwise noted. Littermate controls were used for all experiments. Animal protocols involving these mice were approved by the Animal Studies Committee at Washington University.
Statistical analysis. Data are shown as means ± standard error of the mean (SEM) unless otherwise specified. P values were calculated by the Mann-Whitney, the Student's t-test or the one-way ANOVA, for samples n ¼ 3 and nZ4, respectively, from independent biological replicates and the two-way ANOVA for physiology experiments.
Heatmap analysis. Quantitative reverse transcription polymerase chain reaction (qRT-PCR) data was internally normalized for each genotype and condition to ActB levels. The amount of complimentery DNA (cDNA) used was titrated and verified to yield cycle threshold (CT) values in the linear response range. All conditions where normalized to the values of WT cells at 21% oxygen levels. Resulting data was hierarchically clustered using Cluster 3.0 using the complete linkage cluster method. Java TreeView software generated the colour map.
Hif1 and HIF2 alpha antibodies were validated on heart tissue from embryonic day E11.5 mouse where Hif1 and HIF2 alpha had been deleted. The genotype of the HIF1 and HIF2 conditional KO mice correspond to FGata5 Cre þ Â Hif1a F/ À and FGata5 Cre/ þ Â Hif2a F/ À , respectively. The FGata5 Cre mice have direct Cre recombination in the epicardium as well as in regions of the myocardium 31 .
For histochemistry, hearts were fixed in 4% formaldehyde and processed for paraffin embedding. Sections were stained for Massons' trichrome (collagen stain), PAS (periodic acid acid-Schiff) (to detect glycogen) and Oil red-O (to detect lipid) following standard procedures (Sanford-Burnham-Prebys Histology Core Facility). Sections were mounted and photographed on a TE300 Nikon microscope with 40 Â (N.A. 0.75) objective.
For plastic casting of coronary vascular trees, Microfil polymer (Flow Tech, Inc.) was perfused through the aorta of isolated hearts 2 . Mice had been previously injected with heparin and the heart washed with saline solution to remove blood.  ARTICLE After polymer injection hearts were processed following manufacturer's instructions and imaged on a Nikon SMZ stereo-microscope.
Electron microscopy. For transmitted electron microscopy, hearts were arrested in diastole by perfusion with 300 mM KCl, dissected and sliced in 1 mm thick sections and fixed in 2% pFA/2% glutaraldehyde in 0.1 M cacodylate buffer solution, pH 7.4. Slices were then washed and incubated in 1% osmium tetroxide in the same buffer, dehydrated in ethanol and embedded in Eponate 12 epoxy resin (Ted Pella Inc. C, Redding, CA, USA). Ultrathin sections were prepared on a Reichert-Jung Ultracut microtome and stained with uranyl acetate and lead citrate, and examined using a Hitachi H-600A electron microscope, at 75 kV. Digital images were acquired with 2K Â 2K CCD cooled monochrome 12-bit camera L3C (SIA, Atlanta, GA, USA) using ''Picture Frame'' software (Optronics, Inc., Goleta, CA, USA). Images were processed with ImageJ (public domain software) to achieve a similarity in contrast and brightness of all images.
Imaging quantification. Imaging quantification was done with ImageJ. Colocalization analysis was performed using a colocalization analysis plugin (http://www.uhnres.utoronto.ca/facilities/wcif/imagej/colour_analysis.htm) that generates Mander's coefficients 32 . Determination of distances on images (for example, Z-line and cardiomyocyte cross-sectional area measurements) were done by manually selecting the distances to be measured. For the analysis of the size of ACM in vitro, brightfield images were process on ImageJ as described (http://rsbweb.nih.gov/ij/docs/pdfs/examples.pdf) and quantified using the particle size analysis function.
Intravital endothelial cell labelling. Mice were anesthesized by intraperitoneally (i.p.) ketamine (100 mg kg À 1 )-xylazine (2.5 mg kg À 1 ) injection followed by injection with heparin (1,000 U kg À 1 ) to reduce formation of thrombi. To label the endothelial cells intravitally, the anesthetized mice received a jugular injection of 0.1 ml FITC-conjugated Isolectin B4 (ILB4, Vector Lab). ILB4 was allowed to circulate for 15 min. Mice were then sacrificed and hearts collected in 4% paraformaldehyde, washed in PBS, cryopreserved in sucrose, and embedded in OCT and process for frozen sections.
ACM preparation and culture. ACM were isolated from 3 to5-month-old mice by Langendorff perfusion method as described 33 . Before plating, cardiomyocytes were separated from non-cardiomyocytes by sedimentation. Two hours after plating, adenovirus was applied to the media at an MOI of B50 pfu per cell for 24 h. After changing media on the next day, cardiomyocytes were incubated in a hypoxia chamber (Biospherix XVivo, UK) under specified O 2 tension, as indicated in the text, for 24 h before collection of samples for gene expression analyses or immunodetection.
In situ PLA and immunofluorescence. Isolated ACM seeded on coverslips were fixed on cold methanol for 10 min. After washing with PBS and permeabilization with 1% Triton X-100 cells were process for PLA or immunofluorescence. For immunofluorescence, cardiomyocytes were incubated with blocking buffer (5% BSA, 0.1% Triton X-100) alone for 1 h and then with the primary antibodies overnight at 4°C. Excess of antibody was removed by several PBS washes and alexa fluor-conjugated (488) secondary antibody with rhodamine-conjugated phalloidin (1:1,000) was applied with the blocking buffer for 1 h. After several PBS washes, coverslips were mounted on slides with Vectashield with DAPI (Vector Labs) and photographed.
The proximity ligation assay was performed on coverslips following the Duolink In Situ User Manual (sigma.com/duolink) and using Duolink In Situ Solutions (Sigma). First, we made PLA probes conjugating primary antibodies with a PLUS or MINUS oligonucleotide. We used the Duolink In Situ Probemaker PLUS (Sigma) for RBPJ antibody, RBPSUH (D10A4) XP (Cell Signaling) and MINUS for HIF-1 alpha (NB100-479) and HIF-2 alpha (NB100-122) antibodies. After blocking, cardiomyocytes were incubated with combinations of HIF-1 alpha probe ( À ) and RBPJ probe ( þ ), or HIF-2 alpha probe ( À ) and RBPJ probe ( þ ) overnight at 4°C in a humidified chamber (all 1:250). After washing the antibody, we performed the ligation and amplification steps as described (sigma.com/ duolink) using the Duolink In Situ Detecton Reagents Green (Sigma). After the final washes we incubated cells with rodamine-conjugated phalloiding for 10 min before mounting the coverslips on glass slides using Vectashield with DAPI (Vector Labs).
Endothelial tube formation assay. A microvascular endothelial tube formation assay was performed as described 35 to quantify microvessel formation in response to factors produced by the WT and cKO cardiomyocytes. Briefly, media conditioned by 18 h culture of mouse ACM was collected and centrifuged to remove dead cells and debris. The conditioned media was then applied to human microvascular endothelial (HMEC1) cells plated in 96-well plates previously coated with matrigel (7-8 mg ml À 1 , 45 ml well À 1 ). The cultures were incubated for an additional 16 h either with or without the VEGF receptor inhibitor carbozantinib (Selleckchem). After the 16 h culture period, the angiogenic response was quantified automatically by determining the number of junctions formed using the macro 'Angiogenesis Analyzer' by Gilles Carpentier (Gilles Carpentier. Contribution: Angiogenesis Analyzer, ImageJ News, 5 October 2012) for NIH Image J 1.47v Program.
RNA extraction and qRT-PCR. RNA was extracted from cells and tissue using Trizol reagent (Invitrogen) and cDNA was made using the Qiagen Reverse Transcription kit. PCR with specific primers (see Supplementary Tables 5 and 6) was performed using the LightCycler FastStart Master SYBR Green Kit (Roche) in a LightCycler 2.0 (Roche). Gapdh and b-actin (Actb) were used as internal controls to normalize the data, as specified in the text.
For the angiogenic gene analysis, quantitative Real-Time PCR was performed for the different samples using the iQ SYBR Green Supermix (BioRad) and loaded on a ABI 7900HT. Values are expressed either as 2 DDCt , with DDCt defined as the difference in crossing threshold (Ct) values between experimental and control samples as described 36 or as Log 10 2 DCt using Actb (b-actin) as control gene.
Chromatin immunoprecipitation (ChIP) analysis. ChIP assays were performed using previously described methods 39 . Briefly, ACM from WT and RBPJ cKO mouse hearts were collected, fixed in 2% formaldehyde 12 min at room temperature, rinsed twice in ice-cold PBS, and lysed directly in sonication buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl pH8.0 and protease inhibitors cocktail (Sigma)). The resulting cell suspension was sonicated in a waterbath sonicator (MiSonix Sonicator 3000) using the following programme: 30 s ON, 30 s OFF during10 min (Output Power: 4). The suspension was centrifuged (Eppendorff microfuge) and the cleared supernatant was incubated at 4°C overnight with an antibody to di-acetylated Histone3 (K9, K18) (Millipore, 07-593) or to tri-methylated Histone3 (K4) (Abcam Ab8580) (both 1 mg/10 6 cells). Normal rabbit IgG was the negative control (1 mg/10 6 cells). Protein-DNA complexes were collected on Protein G beads (SIGMA). After rinsing in PBS, the final DNA samples were subjected to PCR. Primer sets specific for the mouse Vegfa, proximal promoters spanning Hif and RBPJ-binding sites were synthesized for this purpose (Supplementary Table 3). PCR conditions were 95°C for 2 min, followed by 25 À 30 cycles at 95°C for 30 s, 55 À 60°C for 30 s and 72°C for 30 s.
Transverse aortic constriction (TAC). Mice between 10 and 14 weeks of age were anesthetized with a ketamine (100 mg kg À 1 )-xylazine (2.5 mg kg À 1 ) mixture administered i.p., and connected to a rodent ventilator after tracheal intubation 40,41 . The chest cavity was opened with scissors by a small incision at the level of the second intercostal space. After isolation of the aortic arch, a 7-0 silk suture was placed around the aorta and a 27-gauge needle. The needle was immediately removed to produce an aorta with a stenotic lumen. The chest cavity was then closed with one 6-0 nylon suture and all layers of muscle and skin closed with 6-0 continuous absorbable and nylon sutures, respectively. Sham-treated animals underwent surgery without the final tightening of the constrictive suture 40,41 . Gradients across the stenotic aorta were measured through echocardiographic analysis, and animals with pressure gradients less than 30 mmHg were not included in the study. Echocardiography was as described 42 . Both males and females were used and no significant gender difference was noted.
Hearts were collected at 14 days after the surgery, arrested in 300 mM KCl and fixed in 4% PFA, or fresh frozen in OCT for histological analysis or, frozen, and stored at À 80°C for RNA and protein extraction.
MI mouse model. The protocol was essentially as described 43 . Briefly, an oblique 8-mm incision was made 2 mm away from the left sternal border toward the left armpit (1-2 mm below). The muscles were separated with care to avoid damaging blood vessels. The chest cavity was then opened at the 4th intercostal space taking care not to damage the lung. A chest retractor was inserted and opened gently to spread the wound to 8-10 mm in width. The pericardium was gently picked up with curved and straight forceps, pulled apart, and placed behind the arms of the retractor. The LAD coronary artery runs in the midst of the heart wall from underneath of the left atrium toward the apex and was typically visible by its pulsating bright red colour. In instances when the LAD artery could not be visualized, the left atrium was lifted so that the source of the LAD artery from the aorta is located. The position of the ligation depends on the volume of infarction desired. In this study, the LAD artery was ligated 1-2 mm below the tip of the left atrium in its normal position, which induces roughly 40-50% ischaemia of the LV. Once the site for ligation had been determined, the curved forceps were used to gently press on the artery a little below the subsequent ligation (to enhance the view of the artery and stabilize the heart). Next, with a tapered needle, a 7-0 silk ligature was passed underneath the LAD coronary artery. For easier and smoother passage the needle was bent in advance to make the curvature rounder. The ligature was then tied with three knots. Occlusion was confirmed by the change of colour (becoming pale) of the anterior wall of the LV. The chest cavity was then closed by bringing together the 4th and the 5th ribs with one or two 6-0 nylon sutures (with pressure applied to the chest wall to reduce the volume of free air). The muscles and skin were closed layer-by-layer with 6-0 absorbable and nylon sutures, respectively. The sham-operated mice underwent the same procedure without tying the suture but moving it behind the LAD artery. Both males and females were used and no significant gender difference was noted.
Animal preparation for cardiac function measurements. Anaesthesia was induced by i.p. injection of sodium pentobarbital (50 mg kg À 1 ) and core body temperature was maintained using a heating pad. Animal preparation included: (i) left femoral artery catheterization, (ii) tracheotomy (polyethylene-90 tube to facilitate spontaneous breathing), and (iii) left ventricle conductance catheter introduction through the right carotid artery. Animals were placed in the supine position on the heating pad, before the experimental procedure. Toe pinching test was performed at least every 5 min, and animals who responded received a small dose of sodium pentobarbital (10 mg kg À 1 ) to prevent animal discomfort. Mice were suitable for the experiments if: (1) systemic parameters were within normal range, namely, heart rate (HR)4300 beat min À 1 , MAP490 mmHg, and arterial O 2 partial pressure (p a O 2 )480 mmHg. Only males were used for these cardiac function measurements. Detailed protocols are described below.
Hypoxia protocol for cardiac function. The anesthetized animals were placed on heating pad within seal acrylic box. As described before for awake animals, the O 2 concentrations in the box was changed using compressed air, 15% O 2 balance N 2 , 10% O 2 balance N 2 ; and 5% O 2 balance N 2 . Each hypoxic step was maintained for 30 min, and measurements were completed 15 min after animal's acclimatization. Hypoxia was stopped if blood pressure dropped below 40 mmHg, and the animal was excluded from the study.
Cardiac function. The closed chest method was used to study cardiac function. Right common carotid artery was exposed to insert a 1.4F pressure-volume conductance catheter (PV catheter; SPR-839, Millar Instruments; Houston, TX, USA). The pressure-volume catheter was advanced passing through the aortic valve into the LV 44 . At baseline and the end of the experiment, a bolus of 15% hypertonic saline (10 ml) was intravenously injected to determine parallel volume 45 . The pressure and volume signals were continuously acquired (MPVS300, Millar Instruments; Houston, TX, USA and PowerLab 8/30, AD Instruments; Colorado Springs, CO, USA). Left ventricular volume was measured continuously in conductance units (RVU; relative volume unit) and converted to actual blood volume (ml) at the end of the experiment.
Cardiac pressure-volume indices. Cardiac function data were analyzed with PVAN software (Millar Instruments, Houston, TX, USA). All cardiac function parameters were averaged from 8 to 12 cardiac cycles at each time point. End systolic pressure (P es ) was directly measured. Cardiac output (CO), stroke work (SW) and stroke volume (SV) were calculated. VR was calculated as using the MAP divided by the CO (VR ¼ MAP/CO). Oxygen delivery (DO 2 ) was calculated as the product of the total Hb by the O 2 carrying capacity of saturated Hb (1.34 mLO 2 / g Hb ) by the arterial blood O 2 saturation and CO (DO 2 ¼ [RBCHb Â 1.34 Â SA] Â CO) 46 .
Hypoxic areas. Hypoxia protocol for awake mice were followed as mentioned earlier. Vital organ hypoxic areas were measured using immunohistochemistry staining for pimonidazole bound to hypoxic zones. Mice received an infusion of the hypoxic marker Hypoxyprobe-1 (pimonidazole 40 mg kg À 1 ) and 5 mg kg À 1 Hoechst 33342 (Invitrogen Corp., Carlsbad, CA, USA) diluted in PBS (total volume, 100 ml), when they were exposed to 10% O 2 . Then, the mice received a second infusion of pimonidazole (40 mg kg À 1 ) and Hoechst 33342 (5 mg kg À 1 ) diluted in PBS (100 ml), when they were exposed 5% O 2 . Finally, mice were euthanized and their heart was removed. Tissues were fixed by immersion in formalin for 24 h at room temperature before transfer to 70% (v/v) ethanol. Lastly, tissues were cut into 100-mm thick sections.
Pimonidazole immunohistochemistry. Sections were cleaned and rehydrated according to standard procedures. Monoclonal antibody directed against pimonidazole (included in the Hypoxyprobe-1 green kit) was used for immunohistochemical staining of the tissue sections. Fluorescence microscopy was performed using an Olympus BX51WI equipped with a high resolution digital CCD ORCA-285 (Hamamatsu Corp., Hamamatsu city, Japan) illuminated with a mercury burner and the appropriate fluorescent cubes (XF100-2 and XF02-2, Omega Optical, Brattleboro, VT). Images for pimonidazole antibody-stained areas and Hoechst were prepared using Wasabi Imaging Software (Hamamatsu Corp). The ratio of pixels stained for pimonidazole in each region, to the total cellular area of the image was calculated. Ten images were analyzed by section, and the results were pooled to determine the mean and standard deviation. To indicate the co-localization of pimonidazole and Hoechst in cell, images were superimposed.
Data availability. All relevant data are included in the figures and/or Supplementary Data, or available from the authors.