YAP and TAZ regulate adherens junction dynamics and endothelial cell distribution during vascular development

Formation of blood vessel networks by sprouting angiogenesis is critical for tissue growth, homeostasis and regeneration. How endothelial cells arise in adequate numbers and arrange suitably to shape functional vascular networks is poorly understood. Here we show that YAP/TAZ promote stretch-induced proliferation and rearrangements of endothelial cells whilst preventing bleeding in developing vessels. Mechanistically, YAP/TAZ increase the turnover of VE-Cadherin and the formation of junction associated intermediate lamellipodia, promoting both cell migration and barrier function maintenance. This is achieved in part by lowering BMP signalling. Consequently, the loss of YAP/TAZ in the mouse leads to stunted sprouting with local aggregation as well as scarcity of endothelial cells, branching irregularities and junction defects. Forced nuclear activity of TAZ instead drives hypersprouting and vascular hyperplasia. We propose a new model in which YAP/TAZ integrate mechanical signals with BMP signaling to maintain junctional compliance and integrity whilst balancing endothelial cell rearrangements in angiogenic vessels.


Introduction
A long-standing question in developmental and cell biology relates to how cells integrate mechanical and chemical signals to orchestrate the morphogenic behaviours that ensure adequate tissue patterning. During sprouting angiogenesis, the arrangement and distribution of cells rather than their numbers appear to drive morphogenesis of the vascular tree. Recent data showing unaltered remodelling in the absence of endothelial cell apoptosis and normal branching frequency across a range of endothelial cell densities support this idea (Watson et al., 2016). In the extreme, however, too few cells will jeopardize network formation and stability (Phng et al., 2009), whereas too many cells might compromise vessel calibre control (Watson et al., 2016). Functional network formation therefore needs to establish the right number of cells in the right place, and distribute them such that the hierarchical branching pattern is supported. What establishes such a balance has remained unclear. Here we provide evidence for the yes-associated protein 1 (YAP) and its paralog WW domain containing transcription regulator 1 (TAZ) as critical endothelial cell autonomous regulators in this process.
YAP and TAZ, two transcriptional co-activators initially discovered as effectors of the Hippo signalling pathway, play a central role in organ size control via regulation of proliferation and apoptosis (Piccolo et al., 2014;Meng et al., 2016;Yu et al., 2015). In confluent cells, YAP and TAZ are phosphorylated by kinases of the Hippo pathway, which induces their retention in the cytoplasm. In sparse cells, YAP and TAZ can translocate to the nucleus, where they interact with transcription factors to regulate the expression of pro-proliferative and anti-apoptotic genes. Other stimuli have been found to regulate YAP and TAZ nuclear translocation and activity -these include, among others, G-protein coupled receptors (GPCRs) , junctional proteins (Giampietro et al., 2015;Schlegelmilch et al., 2011), and mechanical stimuli (Dupont et al., 2011;Aragona et al., 2013). Furthermore, besides cell proliferation and apoptosis, YAP and TAZ also regulate cell differentiation (Yu and Guan, 2013), migration  and actomyosin contraction (Lin et al., 2017). In vascular development, the roles of YAP and TAZ are not fully understood. Yap null mutant zebrafish develop an initially normal vasculature but display increased vessel collapse and regression. Yap/Taz double mutant zebrafish die before the onset of circulation with severe developmental defects, precluding analysis of vascular development in this context (Nakajima et al., 2017). Endothelial-specific deletion of Yap in mice using the Tie2-Cre transgenic line is embryonically lethal due to heart valve defects caused by failed endothelial-to-mesenchymal transition (Zhang et al., 2014). During post-natal development of the mouse retina, YAP was shown to regulate vascular branching and density by promoting the transcription of Angiopoetin-2 (16). While these studies point towards an important role for YAP in regulating blood vessel formation and maintenance, the cellular principles and effectors of YAP/TAZ in endothelial cells in vivo, as well as the possible interplay between YAP/TAZ and the major signalling pathways regulating angiogenesis remain poorly understood.
Here, we used loss and gain of function endothelial specific mouse models to address the roles of YAP and TAZ in the vasculature. We show that YAP and TAZ are both expressed and active in sprouting ECs and critical for sprouting angiogenesis. The inducible, endothelial-specific deletion of YAP and TAZ leads to severe morphogenic defects consistent with impaired junctional remodelling in vivo. We found that the loss of YAP and TAZ decreased VE-Cadherin turnover and decreased the frequency of junction associated intermediate lamellipodia. Furthermore, the loss of YAP and TAZ decreased cell migration and increased cell-cell coupling. We also discovered that endothelial YAP and TAZ strongly inhibit BMP signalling in vitro and in vivo, and that this is mechanistically linked to the migration and permeability defects. Together our results suggest that YAP and TAZ integrate mechanical stimuli with key transcriptional regulators of endothelial sprouting and cell rearrangements during angiogenesis.

YAP and TAZ have distinct expression patterns in endothelial cells of developing vessels and localise to the nucleus at the sprouting front
Immunofluorescence staining in the postnatal mouse retina showed that YAP and TAZ are distinctly expressed in the ECs of the developing vasculature ( Figure 1). While YAP is evenly expressed throughout the vasculature ( Figure 1A-D), the expression of TAZ is especially prominent at the sprouting front ( Figure 1E-H). Furthermore, YAP is exclusively cytoplasmic in all areas of the retinal vasculature, with the exception of the sprouting front where some ECs express nuclear YAP, although at lower levels than in the cytoplasm ( Figure 1A'-D'). TAZ staining signal is very low in the remodelling plexus, arteries and veins ( Figure 1E'-H'); at the sprouting front, TAZ is strongly nuclear in numerous ECs ( Figure 1E, green arrowheads and E'), and both nuclear and cytoplasmic in others ( Figure 1E, red arrowheads). The nuclear signal of YAP and TAZ did not correlate with a tip or stalk cell phenotype; nuclear YAP and TAZ are rather present in a subset of tip and stalk ECs at the sprouting front. YAP and TAZ were also found at endothelial adherens junctions in veins and in the remodelling plexus, (yellow arrowheads in Figure 1D' and F'), as revealed by co-staining for VE- Figure 1. YAP and TAZ are expressed throughout the vasculature of developing mouse retinas, and localise to the nucleus of sprouting endothelial cells. Immunofluorescence staining of YAP (green, A-D and A'-D') and TAZ (green, E-H and E'-H') was performed in wild-type mouse retinas at post-natal day 6 (P6). Retinas were co-stained with the endothelial membrane marker Isolectin-B4 (IB4; blue) and with antibodies against the endothelial nuclei Figure 1 continued on next page YAP/TAZ are required for vascular growth, branching and regularity of the network To examine the cell-autonomous role of endothelial YAP and TAZ during angiogenesis we crossed mice bearing floxed alleles of Yap or Taz (Gruber et al., 2016) with mice expressing a tamoxifeninducible Cre recombinase driven by the endothelial-restricted Pdgfb promoter (Pdgfb-iCreERT2) (Claxton et al., 2008). Injection of the offspring with tamoxifen induced loss of YAP and TAZ protein in ECs during post-natal vascular development, as evidenced by immunofluorescence staining ( Endothelial deletion of YAP or TAZ led to mild vascular defects (Figure 2A,B,C,D). Yap fl/fl Pdgfb-iCreERT2 mice (Yap iEC-KO) presented reduced radial expansion of the vasculature (7% ± 5.4 reduction, p=0.0123) and reduced vessel density (9% ± 4.4 reduction, p=0.0002) ( Figure 2G,H). Taz fl/fl Pdgfb-iCreERT2 mice (Taz iEC-KO) did not show altered radial expansion but displayed decreased vessel density (6% ± 5.8 reduction, p=0.0214) ( Figure 2HG). Neither mutant showed a change in the branching frequency of vessels ( Figure 2I). Interestingly, in Yap iEC-KO retinas the expression of TAZ was increased and TAZ more often localised to the nucleus (Figure 2-figure supplement 2), suggesting compensatory regulation. Taz iEC-KO retinas did not however show a clear difference in YAP expression (data not shown). Deleting both proteins in compound mutant mice (Yap fl/f Taz fl/ fl Pdgfb-iCreERT2, YapTaz iEC-KO) led to a dramatic defect in blood vessel development ( Figure 2E, F): the retinal vasculature showed a 21% (±14, p=0.0012) decrease in radial expansion ( Figure 2G), a 26% (±7.0, p<0.0001) decrease in capillary density ( Figure 2H), and a 55% (±15.4, p<0.0001) decrease in branching frequency ( Figure 2I). Interestingly, the vessel loops were not only bigger in Yap/Taz iEC-KO mice ( Figure 2J), but also more variable in size (Figure 2K), and shape ( Figure 2L) than in control mice. These results indicate that endothelial YAP and TAZ are critical for the development of a homogeneous blood vessel network and can perform redundant functions in the endothelium.
YAP is required for endothelial cell proliferation in response to mechanical stretch As YAP and TAZ display pro-proliferative and anti-apoptotic roles in many cell types (Piccolo et al., 2014;Meng et al., 2016), we evaluated whether the reduced vascularization of Yap/Taz iEC-KO retinas was associated with reduced cell proliferation or increased apoptosis. EC proliferation, assessed by EdU staining ( Figure 3A-C), was decreased in Yap iEC-KO retinas (23% ± 10.0, p=0.0469), whilst not affected in Taz iEC-KO. Consistent with our prior results the decrease in cell proliferation was strongest in Yap/Taz iEC-KO retinas (33% ± 26.0, p=0.0059). Staining for cleaved caspase 3 revealed that apoptosis was unaffected by YAP/TAZ loss ( Figure 3D-F).
To understand if YAP and TAZ were required for proliferation downstream of VEGF, we knocked down YAP and TAZ in human umbilical vein endothelial cells (HUVECs) using small interfering RNAs (siRNAs) (Figure 3-figure supplement 1) and measured the proliferation rate by flow cytometry after treatment with increasing concentrations of VEGF ( Figure 3G). Interestingly, upon loss of YAP, TAZ or YAP/TAZ, ECs proliferated at similar or even increased rates compared to control cells. marker ERG (red). White dotted lines, outline of endothelial nuclei. Yellow dotted lines, outline of perivascular cells' nuclei. Green arrowheads, nuclear localisation of YAP and TAZ. Red arrowheads, cytoplasmic localisation of YAP and TAZ. Yellow arrowheads, junctional localisation of YAP and TAZ. Images correspond to single confocal planes. n > 3 animals for each staining. Scale bar: 10 mm. DOI: https://doi.org/10.7554/eLife.31037.002 The following figure supplement is available for figure 1: Furthermore, VEGF treatment did not alter the subcellular localisation of YAP and TAZ in HUVECs (Figure 3-figure supplement 2), suggesting that VEGF is not a primary regulator of their activity.
We next asked whether YAP and TAZ mediate endothelial proliferation in response to stretchanother crucial mitogenic stimulus for the endothelium (Liu et al., 2007). To this end, we subjected HUVECs to 24 hr of stretch and measured the proliferation rate in comparison to non-stretched, static cells treated with the same siRNAs, by EdU labelling ( Figure 3H). Control cells responded to stretch with a 5-fold average increase in proliferation, and this effect was reduced upon knockdown of VE-Cadherin confirming previous observations (Liu et al., 2007). The knockdown of YAP, but not of TAZ, led to a significant decrease in stretch-induced proliferation. The knockdown of YAP/TAZ showed a tendency to decreased proliferation in response to stretch but did not reach statistical significance. Thus YAP is required for endothelial cell proliferation in response to mechanical stimulation at cell-cell junctions.

YAP/TAZ loss leads to irregular endothelial cell distribution and haemorrhages
Further analysis of Yap/Taz iEC-KO retinas revealed severe defects at the sprouting front. Yap/Taz iEC-KO mutant retinas had 23% (±12.3, p=0.0113) fewer angiogenic sprouts than the control ( Figure 4A,B yellow asterisks and Figure 4-figure supplement 1). Moreover, whereas control sprouts were elongated and showed long cellular protrusions towards the non-vascularised front ( Figure 4A'), sprouts in Yap/Taz iEC-KO retinas were rounder and lacked protrusions ( Figure 4B'). The defective sprout morphology correlated with irregular spacing and frequent aggregations of ECs within the sprouts ( Figure 4B'), arguing that migration and/or the rearrangement of ECs are perturbed in Yap/Taz mutant vessels. Additionally, the Yap/Taz iEC-KO vasculature displayed aberrant vessel crossings ( Figure 4C,C',C",D,D',D"), suggesting that vessels may frequently have failed to anastomose or stabilize connections following sprouting, and instead passed each other. Interestingly, defects in cellular rearrangements, sprouting elongation and anastomosis have previously been associated with altered stability or dynamics of endothelial cell junctions (Sauteur et al., 2014;Bentley et al., 2014;Giannotta et al., 2013;Lenard et al., 2013;Dejana et al., 2008). The defects in vessel morphology were coupled to defects in function as Yap/Taz iEC-KO retinas displayed large haemorrhages from sprouts at the angiogenic front ( Figure 4E,E',F,F'), indicating loss of junctional integrity.
Together, these results argue against the cell proliferation defect being the sole driver of the Yap/Taz iEC-KO phenotype and suggest that endothelial YAP/TAZ play a role in the regulation of EC junctions.

YAP/TAZ regulate adherens junction morphology and stability
Staining for VE-Cadherin revealed several junctional alterations in Yap/Taz iEC-KO vessels ( Figure 4G-H). In control retinas, cell junctions were thin and mostly linear ( Figure 4G'), while in are shown as percentage of the respective controls. Data are mean ±SD. n ! 5 pups. p values were calculated using unpaired t-test. *p<0.05; **p<0.01; ****p<0.0001. (K, L), Quantification of the standard deviation of the area (K) and circularity (L) of the vessels loops in Yap iEC-KO, Taz iEC-KO and YapTaz iEC-KO retinas. Results are shown as percentage of the respective controls. Data are mean ±SD. n ! 5 pups. p values were calculated using unpaired t-test. *p<0.05; **p<0.01; ***p<0.001****p<0.0001. DOI: https://doi.org/10.7554/eLife.31037.004 The following source data and figure supplements are available for figure 2: Source data 1. Values for quantification of radial expansion ( Figure 2G), vessel density ( Figure 2H), branching frequency ( Figure 2I), area of gaps ( Figure 2J) and standard deviation of area ( Figure 2K) and circularity ( Figure 2L   Yap/Taz iEC-KO retinas ECs displayed tortuous junctions ( Figure 4H'). VE-Cadherin staining also unveiled profound differences in the arrangement of ECs within vessels. In control retinas, ECs were   arranged into multicellular tubes, highlighted by the presence of two or more VE-Cadherin junctions running longitudinally along the axis of the vessels ( Figure 4G,G''). Some unicellular segments lacking VE-Cadherin staining could also be found and always correlated with decreasing calibre, indicative of regressing vessels ( Figure 4G red arrowheads) (Franco et al., 2015). In contrast, in Yap/Taz iEC-KO retinas we observed many unicellular vessel segments lacking longitudinal VE-Cadherin junctions, but in vessels of normal calibre ( Figure 4H, red arrowheads and H''). As junctional remodelling has been shown to be required for the cellular rearrangements that establish multicellular tubes (Sauteur et al., 2014), these results suggest that YAP and TAZ regulate junctional remodelling.
VE-Cadherin staining in HUVECs after YAP, TAZ and YAP/TAZ knockdown further revealed altered junctional morphologies compared to control cells. Previous studies have correlated junctional morphology with cellular activities. In vivo, straight or linear junctions were associated with high Notch activity and stalk cell behaviour, while serrated junctions (also referred to as VE-Cadherin fingers) were found in tip cells or actively rearranging cells (Bentley et al., 2014). In vitro, VE-Cadherin fingers were shown to steer migrating ECs and couple leader and follower cells (Hayer et al., 2016), and have also been correlated with increased permeability in cell monolayers. More recently, junction associated intermediate lamellipodia (JAIL) have been identified in the sprouting vessels of the mouse retina, and linked to increased migration (Cao et al., 2017) as well as decreased permeability in cultured ECs (Breslin et al., 2015).
To more accurately describe the differences in junctional morphology after YAP/TAZ knockdown, we defined five junctional categories: straight junctions, thick junctions, thick to reticular junctions, reticular junctions and fingers ( Figure 5E). Live imaging analysis of VE-Cadherin-GFP transduced HUVECs showed that reticular junctions correspond to JAIL, and thick to reticular junctions to small JAIL. ( Figure 5-video 1). Whereas control cells showed mostly reticular junctions ( Figure 5A,F), the knockdown of YAP and TAZ led to an increase in straight junctions and fingers, respectively ( Figure 5B,C,F). The combined knockdown of YAP/TAZ led to an increase in both straight junctions and fingers and to a loss of reticular junctions ( Figure 5D,F). In addition, the knockdown of YAP/TAZ led to junctional breaks in the monolayer, as seen by the presence of gaps in VE-Cadherin stainings ( Figure 5D, red arrowheads). Together, these observations demonstrate that YAP and TAZ together are required for the formation of JAIL and reduce the formation of straight junctions and fingers.
To understand whether this shift in morphology translated into a functional defect, we investigated the permeability of the monolayer to 250 kDa dextran molecules. Only the combined knockdown of YAP/TAZ led to a significant increase in permeability in comparison to the control situation ( Figure 5G), suggesting that YAP/TAZ are both required for the barrier function of the endothelium and can compensate for each other in this particular role.
The dynamic rearrangements of ECs during sprouting require that cell-cell junctions are constantly assembled, rearranged and disassembled. To understand whether YAP and TAZ regulate the turnover of cell junctions, we pulse-labeled VE-Cadherin molecules at cell junctions using an antibody directly coupled to a fluorescent dye for 30 min ( Figure 5H-I) (Dorland et al., 2016). The antibody was subsequently washed out and cells cultured for two more hours in normal conditions, before being fixed and stained for surface VE-Cadherin using a second fluorescent label. Comparing the two sequential VE-cadherin labels allowed us to distinguish junctions with high, intermediate and low turnover rates ( Figure 5J). In control cells, 44% of patches were of high turnover junctions, 24% of intermediate turnover junctions and 32% of low turnover junctions ( Figure 5K). The knockdown of YAP/TAZ significantly decreased the percentage of high turnover junctions to 14% (p=0.0387) and increased the percentage of low turnover junctions to 58%. Interestingly, we found a correlation between the morphology of junctions and VE-Cadherin turnover rates ( Figure 5L): straight junctions and fingers showed the lowest turnover rate, while reticular junctions showed the highest. To understand if the different VE-Cadherin turnover observed after knockdown of YAP/TAZ was caused by a shift in morphology, we compared the turnover of VE-Cadherin within the same morphological categories. Knockdown of YAP/TAZ decreased the percentage of high turnover junctions within all morphological categories, confirming a specific defect in VE-Cadherin turnover. To further investigate how YAP and TAZ affect the turnover of VE-cadherin, we transfected HUVECs with a photo-convertible fluorescent protein tagged VE-Cadherin (VE-cadherin-mEos3.2) and measured the fluorescence loss after photo conversion in straight junctions. The knockdown of YAP/TAZ led to a significant increase in the amount of immobile VE-cadherin-mEos3.2 at the junctions ( Figure 5M) without significantly affecting the half time of fluorescence loss of the mobile fraction ( Figure 5N). Together,  Immunofluorescence staining for f-actin in YAP/TAZ deficient cells ( Figure 5P,P') showed an increase in actin bundles (actin filaments along the junction and stress fibers) and a decrease in branched actin networks, in comparison with control cells ( Figure 5O,O'). This result correlates with the junctional morphologies observed before, as straight junctions and fingers associate respectively with actin filaments and stress fibers (Abu , while JAILs associate with branched actin (Abu . Interestingly, our VE-Cadherin pulse labeling experiments showed that the junction morphologies with less turnover, straight junctions and fingers, were the ones associated with bundled actin filaments, while the morphologies with higher turnover, reticular junctions, associated with branched actin networks. As cell junctions and the actin cytoskeleton are essential for ECs to rearrange and migrate collectively (Vitorino and Meyer, 2008), and Yap/Taz iEC-KO retinas presented less elongated sprouts suggestive of a migration defect, we asked whether cell migration was also regulated by YAP/TAZ.

YAP/TAZ are required for individual endothelial cell migration
To address the requirement of YAP and TAZ for endothelial cell migration, we performed scratchwound assays ( Figure 6A-H). While in the control situation the wound was completely closed at 16 hr ( Figure 6A,B,I), less than 50% of the wound area was closed after YAP knockdown at the same time point (p<0.0001) ( Figure 6C,D,I). A stronger effect on endothelial cell migration was observed after the knockdown of TAZ and YAP/TAZ, with less than 20% of the wound area being closed (p<0.0001 for siTAZ vs siCTR and for siYAP +siTAZ vs siCTR) ( Figure 6E,F,G,H,I).
Given that cells aggregated at the sprouting front of Yap/Taz iEC-KO retinas, we wondered whether in addition to defective directional cell migration they also lacked the ability to shuffle with the neighbouring cells. Recent data illustrated that collectively migrating ECs in vitro move in streams and swirls and display straight junctions along the lateral boundaries and fingers along the front and rear (Hayer et al., 2016). To investigate collective cell migration we therefore analysed the arrangement of cells in a confluent monolayer ( Figure 6J-O). Control cells displayed a cobblestone appearance without identifiable subgroups of cells ( Figure 6J). In contrast, after knockdown of YAP/TAZ cells adopted elongated shapes and arranged into streams and swirls ( Figure 6K). To quantity this effect we used the longest axis of the EC nucleus as a proxy for the orientation of each cell and developed a measure of monolayer coordination based on the alignment of cells with their neighbours ( Figure 6L,L',M,M', N). A score of 1 would signify parallel alignment between all cells, and a score of 0 random alignment of the population. Control cells displayed higher than random alignment with their closest neighbours, but cells beyond 300 mm from each other were arranged at random ( Figure 6I). While the knockdown of YAP did not affect the alignment score of cells, the knockdown of TAZ led to increased alignment. The combined knockdown of YAP/TAZ led to an even higher degree of coordination, with higher alignment scores across all distances between cells. These results suggest that YAP/TAZ promote the ability of cells to distribute individually within monolayers.

Nuclear YAP and TAZ inhibit Notch and BMP signalling in endothelial cells
To gain insight into the nuclear function of YAP and TAZ, we generated a Pdgfb-iCreERT2 -inducible TAZ gain-of-function mouse allele, in which a mutated version of TAZ (TAZ S89A) is introduced in the Rosa26 locus and expressed by a CAG promoter following Cre-mediated excision of an upstream stop cassette (Figure 7-figure supplement 1). The serine-to-alanine mutation in TAZ results in enhanced nuclear TAZ localization as it escapes phosphorylation by the upstream Hippo kinase cascade (Varelas et al., 2010). The allele also expresses nuclear EGFP by means of an IRES sequence, allowing the identification of recombined cells expressing the TAZ mutant protein. Taz iEC-GOF retinas exhibited 25% increased sprouting (±12.2, p=0.0074) ( Figure 7A,B yellow asterisks and Figure 7C) and 19% increased branching (±8.3, p=0.0012) ( Figure 7D). Thus, driving nuclear TAZ expression leads in many aspects to the opposite phenotype of Yap/Taz iEC-KO retinas, suggesting that the loss of the nuclear function of YAP/TAZ plays a key role in the development of the observed vascular loss-of-function phenotypes.
To elucidate the transcriptional targets of YAP and TAZ, we performed unbiased transcriptome analysis on HUVECs transduced with adenoviruses encoding for YAP and TAZ gain-of-function mutants or GFP as a control (AdYAP S127A , AdTAZ S89A , AdGFP) (   Forced activation of YAP and TAZ led to congruent gene expression changes including the canonical YAP/TAZ target genes CYR61, ANKRD1, and CTGF, as expected. Interestingly, YAP and TAZ also suppressed numerous Notch and BMP target genes. During sprouting angiogenesis, Notch and BMP9/10 signalling restrict the acquisition of a tip cell phenotype by activated ECs (Hellströ m et al., 2007;Lobov et al., 2007;Suchting et al., 2007;Siekmann and Lawson, 2007;David et al., 2008;Larrivée et al., 2012;Laux et al., 2013). These results were confirmed by qRT-PCR analysis ( Figure 7E): AdYAP S127A and AdTAZ S89A cells expressed significantly less LFNG, DLL4 and HES1 (Notch target genes), SMAD6, UNC5B and ID1 (BMP target genes) and HEY1 (a common Notch and BMP target gene) than control cells. Consistent with these findings, knockdown of YAP or TAZ lead to a substantial increase in Notch reporter activity (Figure 7-figure supplement 3A) and target gene expression ( Figure 7F). Similar effects were observed for the BMP pathway (Figure 7-figure supplement 3B and Figure 7F), while TEAD-driven reporter activity and YAP/TAZ target genes were repressed (Figure 7-figure supplement 3C and Figure 7F).
To understand if Notch and BMP signalling were also affected in vivo, we stained Yap/Taz iEC-KO retinas for DLL4 and phospho-SMAD1/5/8. In control retinas, DLL4 expression was highest at the leading edge, decreasing over the first 100 mm from the sprouting front, beyond which the expression was evenly low throughout the vessels in the plexus ( Figure 7G,H and Figure 7-figure supplement 4). In Yap/Taz iEC-KO retinas the expression of DLL4 at the sprouting front was higher; additionally, the area of high DLL4 was broader, decreasing for up to 200 mm from the sprouting front before flattening to the lower levels of the plexus. Moreover, staining Yap/Taz iEC-KO retinas for pSMAD1/5/8 showed a~10 fold increase in the number of ECs positive for pSMAD1/5/8 at the sprouting front (p<0.0001) ( Figure 7I,I', J,J',K).
Together, these results identify that endothelial YAP and TAZ repress Notch and BMP signalling during angiogenesis and retinal vascular expansion.

Inhibition of BMP signalling partially rescues the cellular defects of YAP/TAZ loss in endothelial cells
To address whether the Notch or BMP signalling increase were responsible for the cellular defects of YAP/TAZ deficiency, we used different inhibitors to manipulate Notch and BMP signalling in cultured cells. Treatment with the Notch inhibitor DBZ corrected the Notch reporter activity increase after YAP and TAZ knockdown in HUVECs ( Figure 8A), but did not affect the migration of cells in the scratch wound assay ( Figure 8B).
To decrease BMP signaling we screened several inhibitors targeting different BMPs or BMP receptors ( Figure 8C, Supplementary file 1). Only Ldn193187, an Alk2/3/1 kinase inhibitor, significantly decreased the BMP reporter activity after YAP and TAZ knockdown (p=0.0063) ( Figure 8D). In comparison to control cells, the BMP reporter activity in YAP/TAZ deficient cells was 12 times higher, dropping to seven times higher after Ldn193187 treatment. To understand if the partial BMP rescue altered the cellular defects after YAP/TAZ loss, we performed the scratch wound and permeability assays using this inhibitor. Ldn193187 treatment partially rescued the migration defect of YAP/TAZ knockdown HUVECs in the scratch wound assay, significantly increasing the wound closure at 16 hr from 33 ± 3% in DMSO treated cells to 56 ± 5% in Ldn193187 treated cells (p=0.0022) ( Figure 8E). Ldn193187 treatment fully rescued the permeability defect of YAP/TAZ knockdown HUVECs to 250 kDa FITC dextran molecules ( Figure 8F). Staining for VE-Cadherin further showed that Ldn193187 treatment decreased the frequency of finger junctions and increased the frequency of JAIL in YAP/TAZ deficient cells ( Figure 8G-H). Quantification of junction morphologies identified that Ldn193187 treatment decreased the frequency of finger junctions in YAP/TAZ deficient cells to the levels found in control cells ( Figure 8I and Figure 5F, % of fingers: siYAP/TAZ DMSO, 25 ± 5%, siYAP/TAZ Ldn193187 12 ± 0.6%, siCTR: 9 ± 4%). Ldn193187 treatment also led to a 3-fold increase in thick to reticular (p=0.012) and reticular junctions (p=0.043) in YAP/TAZ knockdown cells. The frequency of straight junctions was also significantly decreased (p=0.0326) although still remaining high in comparison to control cells. Together, these results show that the BMP signaling increase is at least partially involved in the cellular defects caused by YAP/TAZ deficiency.
To gain insight into how YAP and TAZ repress BMP signalling, we analysed the gene expression of BMPs, BMP receptors and co-receptors and BMP antagonists in YAP and TAZ gain of function mutant HUVECs (Figure 8-figure supplement 1). As YAP and TAZ function as transcriptional coactivators, the direct targets of YAP and TAZ would be upregulated in this assay, while genes that are downregulated would represent indirect regulation. Forced activation of YAP and TAZ led to increased expression of the BMP ligands BMP2, BMP4 and BMP6, which cannot explain the increase in BMP signalling in the loss of function condition. However, we also found increased expression of the BMP antagonists FST, CTGF, BAMBI, SMURF2, NOG and SMURF1. Thus YAP and TAZ decrease BMP signalling in endothelial cells possibly by increasing the expression of BMP inhibitors.
Together, these results suggest that YAP/TAZ repress BMP activation in endothelial cells, modulating junctions and cell migration.

Discussion
The present study aimed to provide a detailed understanding of the distribution and function of endothelial YAP and TAZ in angiogenesis. Our finding that YAP and TAZ were present in the nucleus of ECs at the sprouting front of developing vessels shows parallels with other cell types where nuclear YAP and TAZ are detected in actively proliferating areas of developing tissues. Interestingly, however, YAP and TAZ show distinct expression patterns in ECs, although these proteins show a high degree of redundancy in many other cell types. While TAZ was predominantly expressed in the sprouting front where it accumulated strongly in endothelial nuclei, YAP was mostly cytoplasmic both in the sprouting front and also in more mature, remodelling vessels. Interestingly, recent work by Sakabe and colleagues showed that cytoplasmic YAP promotes endothelial cell migration (Sakabe et al., 2017).
In addition to nuclear and cytoplasmic YAP and TAZ, we also detected junctional localization of these proteins in retinal vessels. A previous study by Giampietro and colleagues (Giampietro et al., 2015) has shown that endothelial YAP associates with adherens junction proteins at stable junctions and that this prevents its nuclear accumulation and transcriptional activity. Whether this is also true for TAZ has previously not been addressed. A sequestration of YAP and TAZ either in the cytoplasm or bound to junctional proteins can potentially serve different and not necessarily mutually exclusive roles: preventing their nuclear activity, keeping a pool of protein ready to shuttle to the nucleus and drive gene expression, and having other cytoplasmic functions. It is not yet entirely clear what regulates the subcellular localisation of YAP and TAZ in the developing vasculature. ECs at the sprouting front and in more mature vessels have different adherens junctions, experience distinct levels of signalling from secreted angiogenic molecules and are exposed to different levels of shear stress by blood flow. In vitro, endothelial YAP and TAZ relocate to the nucleus upon disruption of cell junctions or loss of VE-Cadherin (shown for YAP by Choi and colleagues (Choi et al., 2015) and confirmed in our analysis also for TAZ, data not shown). In the mouse retina vasculature, Cao and colleagues (Cao et al., 2017) recently observed that endothelial cells at the sprouting front display reduced relative VE-Cadherin concentration that promotes cell-cell junction dynamics and JAIL formation. These observations nicely correlate with our own data showing the sprouting front as the area of preferential nuclear YAP/TAZ. Furthermore, we did not find junctional localisation of YAP/ TAZ at the sprouting front, pointing to less sequestration of YAP/TAZ away from the nucleus by more dynamics junctions. Together, these data provide additional strength to a model in which reduced VE-Cadherin concentration at cell junctions may promote YAP and TAZ relocation to the nucleus. In zebrafish, Nakajima and colleagues (Nakajima et al., 2017) showed that YAP nuclear relocation correlated with lumenisation of sprouting vessels, and they attributed this to the effect of shear stress on YAP. In the mouse retina, hemodynamic fluid laws predict that vessels at the sprouting front experience very low levels of shear (Bernabeu et al., 2014), arguing against YAP and TAZ being activated by shear in this model. Additionally, we found no difference in the subcellular localisation of YAP or TAZ between arteries and veins, that is, vessels that experience distinct shear stress levels. However, it is possible that local and fast changes in shear stress levels are more relevant to regulate YAP and TAZ than sustained shear. In support of this idea, YAP and TAZ appear not to respond to 12 or 24 hr of laminar shear , but translocate to the nucleus after only 10 min of laminar shear (Nakajima et al., 2017). Finally, although VEGF, a pro-angiogenic molecule secreted by astrocytes at the avascular front, drives endothelial proliferation and migration, we found no evidence for VEGF induced YAP and TAZ nuclear translocation, although others found opposite results (Kim et al., 2017;Wang et al., 2017). Other pro-angiogenic molecules, either locally produced or blood-borne, could regulate endothelial YAP and TAZ during development; future work will help clarify these questions and how different chemical and mechanical stimuli come together to regulate YAP and TAZ.
To address the cell autonomous role of YAP and TAZ we took advantage of an endothelial specific inducible Cre to inactivate YAP and/or TAZ genetically during angiogenesis. The mild phenotype of the single mutants in comparison to the drastic phenotype of the compound mutant indicates functional redundancy in the endothelium. The compound loss of endothelial YAP and TAZ leads in the mouse retina to a decrease in the radial expansion of vessels, vascular density, branching and sprouting. This phenotype could be a consequence of a decreased number of ECs caused by a proliferation defect (Ubezio et al., 2016). However, our further discovery that YAP and TAZ are required to establish homogeneity in the plexus and prevent cellular aggregations suggests that endothelial YAP/TAZ signalling is not only required to provide adequate numbers of cells but is also critically involved in ensuring adequate EC distribution. We propose that endothelial YAP/TAZ operate in several mechanisms that jointly establish a balance of the right number of endothelial cells in the right place. First, endothelial YAP/TAZ drive proliferation in response to mechanical stimulation at the cell-cell junction, and not in response to VEGF. We propose that in this way endothelial YAP/ TAZ provide a cell intrinsic mechanism of locally controlling cell densities, in contrast to growth factor-mediated cell proliferation instructed by the surrounding tissue. Second, endothelial YAP/TAZ increase VE-Cadherin turnover at cell-cell junctions, which we propose is essential for cells to migrate and rearrange while maintaining the endothelial barrier at the same time. This corroborates recent findings in mouse hepatocytes where YAP antagonises adherens junction stability (Bai et al., 2016). The authors showed that YAP regulates hepatocyte adherens junctions in response to increased actomyosin contractility by increasing myosin II light chain gene expression. Accordingly, the transcriptional, nuclear role of YAP was required for junctional regulation. Together with our observations, these findings indicate the existence of a positive feedback loop where stable junctions sequester YAP and TAZ from the nucleus, therefore maintaining less junctional turnover, while remodelling junctions allow YAP and TAZ to relocate to the nucleus where they increase VE-Cadherin turnover. Our results also suggest that a high VE-Cadherin turnover at the sprouting front is required in order to maintain junctional integrity and prevent bleedings. YAP and TAZ increase the presence of JAIL (the more dynamic type of junction in our analysis) and promote branching of actin. These results are in accordance with recent data from Sakabe and colleagues reporting that YAP and TAZ increase Cdc42 activity in lamellipodia and phosphorylation of N-WASP (Sakabe et al., 2017), an actin binding protein that promotes branching of actin through the activation of the Arp2/ 3 complex.
Molecularly, how YAP and TAZ affect this complex cell behavior is not entirely clear. Our results identify that endothelial YAP/TAZ reduce the expression of Notch and BMP signaling in ECs, yet the in vitro rescue attempts show that the Notch increase is not the reason for the YAP/TAZ phenotype. This is further corroborated by in vivo data from Kim and colleagues showing that DAPT treatment fails to restore sprouting defects in YAP/TAZ mutant retinas (Kim et al., 2017). In contrast, reducing the increased BMP signalling experimentally corrected the permeability, and partially restored cell migration in YAP/TAZ deficient cells. This would suggest that BMP, not Notch, is a main driver of the observed cellular phenotypes. Whether the increased BMP signalling is also responsible for the phenotypes of YAP/TAZ loss of function in vivo remains to be shown. Furthermore, the full details of the involved ligands and receptors of the BMP pathway remain to be determined. Previous studies identified that BMP9/10 -Alk1 signalling is anti-angiogenic (Larrivée et al., 2012), while BMP2/4/6 -Alk2/Alk3 signalling is pro-angiogenic . Given the hyposprouting phenotype and reduced cell migration, we expected an increase in BMP9/10-Alk1 signalling after YAP/TAZ loss. However, the results from the panel of BMP inhibitors instead point towards a possible increase of the BMP2/4 -ALK3 pathway activity. However, given the notorious promiscuity of chemical inhibitors, and the different cellular context in which they are tested, this result should be seen as an indication at best. Further studies will need to address the nature of the BMP ligand/receptors and how the deregulated BMP signalling affects endothelial cell migration and adherens junctions in the context of YAP/TAZ deficiency.
Together, our results identify a role for YAP/TAZ in promoting endothelial cellular rearrangements through the regulation of junctional turnover and collectiveness of cell migration. Conceptually, linking stretch-induced proliferation (to balance cell numbers) with modulation of junctional turnover (to facilitate cell rearrangements) seems ideally suited to achieve the required balance of cell distribution for functional vascular patterning.

Mice and treatments
For loss of function experiments the following mouse strains were used: Yap fl/fl and Taz fl/fl (Gruber et al., 2016), Pdgfb-iCreERT2 (Claxton et al., 2008). A detailed description of the knock-in mice overexpressing the TAZ gain-of-function allele will be provided elsewhere. Briefly, 3xFLAG-TAZ S89A -IRES-nEGFP with a preceding floxed STOP cassette was knocked into the Rosa26 locus. Cre-mediated removal of the STOP sequence leads to CAG promoter-driven expression of 3xFLAGtagged TAZ S89A as well as of nuclear-localized enhanced green fluorescence protein (nEGFP). The allele was kept heterozygous in the experimental studies and was developed together with genOway.
Mice were maintained at the London Research Institute and at the Max Delbruck Center for Molecular Medicine (loss of function mice) and at the Max Planck Institute for Heart and Lung Research (gain of function mice) under standard husbandry conditions. To induce Cre-mediated recombination 4-hydroxytamoxifen (Sigma, 7904) was injected intraperitoneally (IP) (20 mL/g of 1 mg/mL solution) at postnatal day 1 and day 3 and eyes were collected at P6. In all loss and gain of function experiments control animals were littermate animals without Cre expression. Male and female mice were used for the analysis.
For endothelial cell proliferation assessment in the retina, mouse pups were injected IP 2 hours before culling with 20 uL/g of EdU solution (0.5 mg/mL; Thermo Fischer Scientific, C10340).

Cell culture
HUVECs from pooled donors (PromoCell) were cultured in EGM2-Bulletkit without antibiotics (Lonza) and used until passage 6. For YAP and TAZ gain of function experiments HUVECs were obtained from Lonza, cultured in endothelial basal medium (Lonza) supplemented with hydrocortisone (1 mg ml À1 ), bovine brain extract (12 mg ml À1 ), gentamicin (50 mg mlÀ1), amphotericin B (50 ng ml À1 ), epidermal growth factor (10 ng ml À1 ) and 10% fetal bovine serum (Life Technologies) and used until passage 4. The manufacturers authenticated the identity of HUVECs by flow cytometry for cell-type specific markers (vWF, CD31, CD105) and by functional analysis (cells positive for acetylated low density lipoprotein uptake). All cells were tested negative for mycoplasma.
To activate YAP and TAZ signalling in ECs, FLAG-YAP S127A -or 3x-FLAG-TAZ S89A -encoding adenoviruses were generated in the adenoviral type five backbone lacking the E1/E3 genes (Vector Biolabs). GFP-encoding adenoviruses were used as a control. Infections were carried out by incubating sub-confluent HUVECs (70-80%) with starvation media (EBM containing 0.1% BSA) for 4 hr followed by the addition of adenoviral particles and polybrene (Santa Cruz). After 4 hr, HUVECs were washed with Hanks Buffer for at least five times and then cultured in complete EBM media with 10% FCS and supplements overnight. All experiments were performed 24 hr post transduction.

Immunofluorescence staining
To perform retina immunofluorescence, eyes were collected from postnatal day six mice and fixed in 4% PFA in PBS for 1 hr at 4C. Retinas were dissected in PBS and permeabilised/blocked for 1 hr at room temperature in 1% BSA, 2% FBS, 0.5% Triton X100, 0.01% Na deoxycholate and 0,02% Na Azide in PBS. Primary and secondary antibodies were incubated overnight at 4C and for 2 hr at room temperature, respectively, both in 1:1 PBS: blocking buffer. Isolectin staining was performed overnight at 4C in Pblec after retinas were equilibrated for 1 hr in Pblec at room temperature. Retinas were post-stained fixed in 2% PFA in PBS for 10 min. To mount the samples Vectashield mounting medium. (Vector Labs, H1000) or ProLong Gold (Thermo Fisher Scientific) was used. Imaging was done by laser scanning confocal microscopy (Carl Zeiss LSM700, LSM780 and Leica TCS SP8). Processing of samples was carried out in tissues from littermates under the same conditions.
For immunofluorescence in HUVECs, cells were grown in #1.5 coverslips coated with poly-lysine and gelatin 0.2%. At the end of the experiment cells were fixed in 4% PFA for 10 min, permeabilised in 0.3% Triton-X100 in blocking buffer for 5 min and blocked in 1% BSA 20 mM Glycine in PBS for 30 min. Primary and secondary antibodies were incubated for 2 and 1 hr, respectively, in blocking buffer. Nuclei labeling was performed by incubating cells with DAPI for 5 min (Life technologies, D1306).
A list of the primary antibodies used can be found in Supplementary file 2.

Image analysis
Analysis of radial expansion, capillary density, branching frequency, proliferating ECs, apoptosis and sprouting numbers was done using Fiji (Schindelin et al., 2012). Radial expansion corresponds to the mean distance from the optic nerve to the sprouting front (eight measurements in tilescans of two whole retinas per animal). Capillary density corresponds to the vessel area (measured by thresholding IB4 signal) divided by the field of view area (6-8 images of (425 mm) 2 between artery and vein per animal). Branching frequency was measured by manually counting all branching points in a field of view (4-5 images of (200 mm) 2 between artery and vein per animal). The plexus regularity was assessed through the standard deviation of the size and the circularity of the vascular loops in the plexus (using same images as for analysis of capillary density). Vascular loops were segmented by thresholding the IB4 signal to avoid artifacts we excluded loops with a size smaller than 86 um^2 for the analysis. Endothelial proliferation was measured by manually counting the number of EdU positive endothelial nuclei (ERG positive) and dividing by the vessel area (measured by thresholding IB4 signal) (4 images of (425 mm) 2 containing the sprouting front and localized on top of arteries per animal). Apoptosis was measured manually by counting the number of cleaved caspase 3 positive figures and dividing by the vessel area (measured by thresholding IB4 signal) (tilescan of one whole retina per animal). The number of sprouts was measured manually (3 images of 425 Â 850 mm of the sprouting front per animal). To quantify DLL4 intensity the outline of the sprouting front and the position of the arteries were manually defined using IB4 staining. Vessels were segmented by thresholding the IB4 staining in Fiji. Then, DLL4 intensity inside the vasculature was normalised with the average DLL4 intensity outside of the vasculature. Subsequently, for every pixel inside the vasculature (excluding the arteries) the distance to the sprouting front was calculated. The normalised DLL4 values within each bin were averaged (15 mm bins from 0 to 500 mm). For each retina quarter a curve was obtained, and the average and SEM of these curves was shown in the graph (one retina quarter was used per animal). To quantify pSMAD1/5/8 status the number of pSMAD1/5/8 positive endothelial nuclei was manually counted and dividing by the total number of endothelial nuclei (defined by being ERG positive) (3 images of (225 mm) 2 containing the sprouting front were used per animal).
To analyse YAP/TAZ subcellular localisation in HUVECs we adapted a previously existing cytoplasm-to-nucleus translocation assay pipeline from Cell Profiler (Carpenter et al., 2006). Briefly, YAP or TAZ staining intensity was measured both inside the nucleus of the cell and in a 12 pixels wide ring of cytoplasm grown radially from the nucleus. The nucleus localisation was determined using a DAPI mask.
Cell junction morphology analysis was done in confluent monolayers of HUVECs stained for VE-Cadherin. Five morphological categories were defined: straight, thick, thick to reticular, reticular and fingers. We acquired 5 images of (160 mm) 2 per condition per experiment, divided each image in (16 mm) 2 patches, and randomly grouped these patches. The classification into categories was done manually and blindly for the condition.
To analyse cell coordination we used confluent cells labelled for DAPI. The nuclei were automatically segmented using a customized Python algorithm relying on the Scikit Image Library. By fitting an ellipse to each nucleus we obtained its major and minor axis, and the angle of the major axis with the x-axis of the image was assigned to the nucleus as its orientation. This way each nucleus in the images was assigned a position given by its midpoint and an orientation. Next we analyzed the average alignment of the nuclei of two cells depending on their distance. As the nuclei don't have a directionality (i.e. they are nematics as opposed to vectors), the angles between two nuclei range from 0 corresponding to the nuclei being parallel, to p/2 corresponding to them spanning a right angle. For any two cells in each image we calculated the angle and the Euclidean distance between them, and then we binned the cells depending on their distance. We introduced a parameter called 'alignment' which is one if all cells are perfectly aligned and 0 for a completely random distribution of cell orientations.

Live imaging of VE-Cadherin-EGFP
24 hr after siRNA transfection, knockdown HUVECs were transduced with VE-cadherin-EGFP adenovirus as described before (Bentley et al., 2014). Briefly, cells were incubated with the virus for 24 hr and then washed three times to remove viral particles. Cells were replated onto 2-well LabTek chambered coverslips (Nunc) coated with 10 ug/mL Fibronectin (Sigma, F1141). Imaging was performed 48 hr post-transduction. Cells were imaged at 37˚C under 5% CO 2 on LSM 780 (Zeiss) using a Plan-Apochromat 63x/1.4 oil objective. Images were acquired at a 260 s time frame.
VEGF treatment and YAP/TAZ staining Confluent HUVECs were maintained in VEGF free media for 24 hr. VEGF treatment was then performed for 30 min, 1 hr and 3 hr with 0 or 40 ng/mL of .
Immunofluorescence staining and analysis of YAP and TAZ subcellular localisation was performed as above described.

VEGF treatment and proliferation assessment
Knockdown HUVECs were maintained in VEGF free media for 24 hr. VEGF treatment was then performed for 24 hr with 0 ng/mL, 40 ng/mL, 200 ng/mL or 1000 ng/mL of VEGF-165. Cells were pelleted, ressuspended in 90% cold Methanol and stored at À20C˚before further processing. Cells were then ressuspended in Propidium Iodide/RNase staining solution (Cell signaling, 4087) for 30 min before cell cycle analysis by flow cytometry (LSRII, BD). Data was analysed using BD FACSDiva software.

Mechanical stretch application and proliferation assessment
HUVECs were plated on collagen I -0.2% gelatine-coated Bioflex plates (BF-3001C, Flexcell International Corporation). Gene knockdown was preformed as previously described. Cells were incubated in transfection media for 24 hr, and allowed to recover in fresh complete media for 4 hr. Afterwards cells were incubated for 24 hr in serum starvation media (0,1%BSA in EBM2 pure media) to form a confluent, quiescent monolayer. Cyclic stretch (0.25 Hz, 15% elongation) was then applied for 24 hr using a Flexcell FX-5000 Tension System. Control cells were placed in the same incubator but not on the Flexcell device (static conditions). EdU pulsing was performed after 20 hr of the 24 hr stretch period. At the end of the experiment cells were fixed in 4% PFA and EdU staining was performed according to the manufacturer's protocol (Click-It EdU C10340 Life Technologies). Nuclei were labelled with DAPI. Three regions of interested were acquired per sample in a Carl Zeiss LSM700 scanning confocal microscopes (Zeiss, Germany). Quantification of proliferation was done using a CellProfiler pipeline. Percentage of S phase cells was determined as percentage of EdU positive nuclei over the total number of nuclei.
Permeability assay 24 hr after siRNA transfection cells were re-plated into fibronectin coated Transwell membranes (Costar 3460) at confluence and incubated for two more days to stabilize cell junctions. On the third day after transfection 0.5 mg/mL of 250 kDa FITC Dextran in cell media (Sigma FD250) was added to the top well. Fluorescence on the bottom well was measured after 6 hr in a Gemini XPS fluorescent plate reader.

Pulse chase VE-Cadherin experiment for quantification of low, intermediate and high turnover junctions
Cells were labelled live with a non-blocking monoclonal antibody directed against extracellular VE-Cadherin and directly coupled with Alexa-Fluor647 (BD Pharmingen, #561567, 1:200) for 30 min. Cells were then washed 2x with PBS and incubated with complete media for additional 2 hr. Cells were fixed with 4% PFA and stained for VE-Cadherin (Santa Cruz Biotechnology, #6458, 1:200) with a secondary antibody coupled with Alexa-Fluor-488. 5 (160mm) 2 images per condition per experiment were acquired in a Carl Zeiss LSM700 confocal laser scanning microscope using the same acquisition settings. Max projection of z stack and merging of channels was done in Fiji. Images were divided in (16 mm) 2 patches and the patches were randomly grouped. Patches were classified into a morphological category and into low, intermediate or high turnover categories, manually and blindly for the condition.
Cells were imaged at 37˚C under 5% CO 2 on LSM 780 (Zeiss) equipped with Definite Focus stabilizer. Imaging was performed using the 488 nm (green mEos3.2 component) and the 561 nm (red component) lasers using Plan-Apochromat 63x/1.4 oil objective, 0.26 Â 0.26 mm pixel size and 5.09 ms pixel dwell time, 16-bit image depth. A circular region of interest (ROI) of 21 mm 2 area was selected on straight junctions and photoconverted using the 405 nm laser. Mean fluorescence intensity in the ROI was monitored in the red channel for 15 min with 10 s resolution, while the movement of the junction was followed in the green channel.
Background signal in each frame was estimated by measuring mean intensity in non-photoconverted region and subtracted from the fluorescence-loss curves. The curves were further corrected for bleaching, using parameters estimated from fixed cells. The curves were normalized between the mean intensity in the frames before photoconversion and fluorescence intensity measured immediately after photoconversion. Normalized curves were smoothed using the moving average method; the half-time of redistribution and immobile fraction of VE-cadherin were estimated directly from the plots. Only junctions which did not substantially move or remodel during the observation time were considered for analysis. Analysis was performed using Fiji and Matlab (Mathworks).

Scratch wound assay
24 hr after siRNA transfection cells were re-plated into a scratch wound assay device (IBIDI). On the following day a cell free gap of 500 mm was created by removing the insert of the device. Images were taken immediately after removing the insert (0 hr) and after 16 hr. The cell free area was measured in Fiji and used to calculate the percentage of wound closure at 16 hr.

RNA extraction and quantitative real time-polymerase chain reaction
RNA was extracted using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. For HUVECs transfected with adenoviruses carrying YAP and TAZ gain of function mutations, 2 mg of total RNA were reverse transcribed to cDNA using M-MLV reverse transcriptase (ThermoFisher Scientific). For HUVECs transfected with siRNAs 90 ng of RNA were reverse transcribed using RevertAid First Strand cDNA Synthesis Kit (ThermoFisher Scientific). qRT-PCR was performed using TaqMan reagents and probes (Applied Biosystems) (listed in Supplementary file 3). qRT-PCR reactions were run on a StepOnePlus real-time PCR instrument (ThermoFisher Scientific) or Quant Studio 6 Flex (Applied Biosystems) and expression levels were normalised to human ACTB or human HPRT1 using the 2deltaCT method.

Western blot
Protein was extracted from HUVECs using M-PER protein extraction reagent with Halt Protease and Phosphatase inhibitors (Pierce). Proteins concentration was assessed using a BCA protein assay kit (Pierce). Proteins were separated by SDS-PAGE and blotted onto nitrocellulose membranes (Bio-Rad). Membranes were probed with specific primary antibodies and then with peroxidase-conjugated secondary antibodies. The following antibodies were used: YAP 63.7 (Santa Cruz Biotechnology, sc-101199, 1:1000), GAPDH (Millipore, MAB374, 1:4000). The bands were visualized by chemiluminescence using an ECL detection kit (GE Heathcare) and a My ECL Imager (Thermo Scientific).

Dual luciferase reporter assay
Renilla-luciferase reporter assays for TEF-1 (Mahoney et al., 2005), RBPj (Jarriault et al., 1995), BRE (Korchynskyi and ten Dijke, 2002;Fritzmann et al., 2009) and FOPflash (Korinek et al., 1997)-Luciferase promoter activity were performed as follows: 48 hr after gene knockdown by siRNA HUVECs were cotransfected with 600 ng of Luciferase reporter gene construct and 300 ng of pRL-TK (Promega) using Lipofectamine2000 and incubated for 4 hr. Cell extracts were prepared 72 hr post siRNA transfection and 24 hr post Luciferase reporter transfection, and luciferase activity was measured using a dual luciferase system as described (Hampf and Gossen, 2006). Experiments were carried out in duplicates and results were normalized to the correspondent FOPflash/Renilla measurement.

Microarray and gene set enrichment analysis
Microarray studies were performed as described (Murtaugh et al., 2003). In brief, total RNA was extracted from HUVECs using the RNeasy kit (Qiagen) and RNA quality assessed with the 6000 nano kit and an Agilent Bioanalyser. RNA was labelled according to the Affymetrix Whole Transcript Sense Target Labeling protocol. Affymetrix GeneChip Human Gene 2.0 ST arrays were hybridized and scanned using Affymetrix protocols. Data were analysed using the Affymetrix expression console using the RMA algorithm; statistical analysis was done using DNAStar Arraystar 11. Heat maps of gene signatures were plotted using RStudio, Inc.

Notch and BMP inhibition experiments
A list of the reagents used, together with duration of treatment, can be found in Supplementary file 1.

Statistical analysis
Statistical analyses were performed using GraphPad Prism software and p value was determined using unpaired Student t-test between the control and the knockout/knockdown/ gain of function condition. Statistical significance was considered for p<0.05. Values shown are mean and standard deviation was used as the dispersion measure. Biological replicates refer to individual mice for in vivo experiments and different wells for in vitro cell culture experiments; independent experiments refer to experiments done in different days; technical replicates refer to repeated measurements taken from the same sample, both for in vivo and in vitro. Exclusion of outliers was done using 'Robust regression and Outlier removal' from GraphPad Prism software, with a coefficient Q of 1%. A statistical method of sample size calculation was not used during study design. For in vivo experiments, we used an average of 6 animals per experiment, from different litters, with a minimum of 3 (detailed number of animals used in figure legends and source data). For in vitro experiments, we did a minimum of 3 independent experiments (detailed number of experiments in figure legends and source data). When technically possible the investigators were blind to the genotype of the animal or cell culture condition during sample processing and data analysis. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.