YES kinase controls endothelial junctional plasticity and barrier integrity by regulating VE-cadherin phosphorylation and endocytosis

Vascular Endothelial (VE)-cadherin in endothelial adherens junctions is an essential component of the vascular barrier, critical for tissue homeostasis and implicated in progression of diseases such as cancer and eye diseases. Inhibitors of SRC cytoplasmic tyrosine kinase have been applied to suppress tyrosine phosphorylation of VE-cadherin and thereby to prevent excessive leakage, edema and high interstitial pressure. We show that the SRC-related YES tyrosine kinase rather than SRC, is localized at endothelial cell (EC) junctions. EC-specic YES deletion suppresses VE-cadherin phosphorylation, and arrests VE-cadherin at EC junctions. This is accompanied by loss of EC collective migration, and exaggerated agonist-induced macromolecular leakage, while extravasation of monocytes is suppressed. Overexpression of Yes causes ectopic VE-cadherin phosphorylation while vascular leakage is unaffected. In contrast, in EC-specic Src-decient mice, VE-cadherin internalization is maintained and leakage is suppressed. In conclusion, YES-mediated VE-cadherin phosphorylation regulates its constitutive turnover, required for endothelial junction plasticity and vascular integrity. proteins were denatured and released from the beads by boiling in β-Mercaptoethanol-containing Laemmli buffer. Western blot. Protein samples were separated on a SDS-PAGE (4-12% gradient gel) (ThermoFisher Scientic), transferred to nitrocellulose membranes (GE Healthcare), and incubated sequentially with primary and appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies. Signals were detected using the ECL reagent kit (Cytiva). Pictures were obtained by BioRad ChemiDocMP imaging system and analyzed using Image Lab software. The following primary antibodies were used: rabbit anti-VE-cadherin Y685 (CP1981, ECM Biosciences, 1:1000); goat anti mouse VE-cadherin (AF1002, R&D Systems, 1:1000); rabbit anti-pSRC (Y418) (44-660G, ThermoFisher Scientic, 1:1000); mouse anti-YES (610376, BD Biosciences, 1:1000); mouse anti-p120-Catenin (610133, BD Biosciences, 1:1000). Secondary antibodies used were: Horse radish peroxidase (HRP)-conjugated anti-rabbit IgG (NA934, Cytiva); HRP-conjugated anti-mouse IgG (NA931, Cytiva); HRP-conjugated anti-goat IgG (P0449, Dako). Secondary antibodies were used at a dilution of 1:7500.

Overexpression of Yes causes ectopic VE-cadherin phosphorylation while vascular leakage is unaffected.
In contrast, in EC-speci c Src-de cient mice, VE-cadherin internalization is maintained and leakage is suppressed. In conclusion, YES-mediated VE-cadherin phosphorylation regulates its constitutive turnover, required for endothelial junction plasticity and vascular integrity.

Main
Endothelial cell-cell adhesions form an integral part of the vascular barrier, restricting the passage of molecules and cells across the vessel wall 1 . The barrier in most organs is exible, allowing remodeling of the embryonic and postnatal vascular beds and exchange of small molecules and uid to maintain adult tissue homeostasis 2 . During acute in ammation and in a range of chronic diseases, the barrier is weakened, leading to extravasation of blood constituents, promoting in ammation and progression of pathologies such as retinopathies and cancer [1][2][3] . The maintenance of junctions and barrier function relies on junctional complexes of cell-adhesion molecules formed between adjacent endothelial cells (ECs).
Tight junctions limit vascular leakage in the central nervous system (CNS) as well as in arteries and postarterial capillaries in peripheral tissues 4,5 . Adherens junctions, formed by vascular endothelial (VE)cadherin and associated catenins, are crucial for regulating venous and capillary vascular permeability, and leukocyte extravasation as well as collective cell migration 6-12 . According to the consensus model, VE-cadherin is rapidly internalized in response to in ammatory cytokines and vascular endothelial growth factor A (VEGFA), allowing transient opening of gaps at EC contacts 1 . Although tyrosine phosphorylation of VE-cadherin has been implicated in regulation of its internalization, the mechanisms in gap formation and vascular leakage have remained poorly understood 13,14 . This is important, as exaggerated vascular leakage and formation of edema is a driver of disease 1 .
Three main tyrosine phosphorylation sites have been identi ed in VE-cadherin; Y658, Y685 and Y731 15,16 , which have been implicated in distinct processes. Both Y658 and Y685 are phosphorylated in veins, but not in arteries, as a consequence of ow-induced activation of SRC cytoplasmic tyrosine kinases 17 . Phosphorylated (p)Y658 relieves tension across VE-cadherin through binding of the polarity protein LGN (Leucine-Glycine-Asparagine repeat), which displaces the VE-cadherin-associated p120 catenin 18 .
Phosphorylation of Y685 is induced by ow as well as by in ammatory cytokines and VEGFA 16,17,19 .
Vascular leakage requires phosphorylation on Y685 as inferred from a Y685-to phenylalanine (F) exchange mutant mouse, which shows suppressed extravasation of large molecular weight tracers or cells in agonist-treated healthy tissues, in eye disease and in ammatory models 10,20,21 . Agonist-induced leakage is accompanied by reduced levels of pVE-cadherin, at least in part due to VE-cadherin endocytosis 17 . Y731 is constitutively phosphorylated already upon VE-cadherin synthesis and its dephosphorylation by Src homology phosphatase 2 (SHP2) is a prerequisite for leukocyte extravasation 10,22 . Several additional phosphotyrosine phosphatases (PTPs) such as VE-PTP and density enhanced phosphatase 1 (DEP1; also denoted CD148) have been implicated in regulation of VEcadherin phosphorylation 23,24,25 .
In order to stabilize the vascular barrier in disease, a range of inhibitors have been developed to suppress the activity of SRC, which has been identi ed as the upstream kinase responsible for the phosphorylation of VE-cadherin on Y658 and Y685 16,17 . SRC-mediated VE-cadherin phosphorylation on Y685 in response to VEGFA stimulation has been regarded as an important step in VEGFA induced leakage 10,16,18 .
However, EC-speci c Src knockout mice do not display signi cant changes in VE-cadherin phosphorylation 26 . In addition to SRC, ECs express the highly related cytoplasmic tyrosine kinases YES and FYN. While SRC is implicated in regulation of cell-matrix adhesion 26,27 , the roles played by endothelial YES and FYN have remained poorly understood. In vitro studies have suggested distinct roles for SRC, YES and FYN in regulating EC behaviors, despite their well-conserved SRC Homology 2 (SH2), SH3 and kinase domains 28 . Accordingly, constitutive global gene inactivation results in abolished VEGFAinduced vascular leakage in constitutive gene inactivated Src -/or Yes -/mice, but not in Fyn -/mice 29 .
Here, using inducible EC-speci c knockout and overexpression mouse models we demonstrate a critical role for YES in the phosphorylation of VE-cadherin at all main phosphorylation sites, Y658, Y685 and Y731. By comparing the effects of losing Src or Yes on VE-cadherin phosphorylation in vivo and in vitro, we conclude that YES is the main kinase that maintains phosphorylation of a subset, about 30%, of the VE-cadherin pool in veins and capillaries, normally leaving most of the pool unphosphorylated. Loss of VE-cadherin phosphorylation suppresses its endocytosis and leads to abnormal barrier properties with blunted collective migration, excessive macromolecular leakage and suppressed leukocyte extravasation. However, forced phosphorylation of VE-cadherin by Yes overexpression does not affect agonist-induced leakage; thus, macromolecular leakage is not controlled solely by the extent of VEcadherin phosphorylation. In contrast, in Src-de ciency, VE-cadherin endocytosis is unaffected and vascular leakage is reduced. Therefore, YES and SRC differently regulate VE-cadherin turnover and adherens junction stability. We conclude that therapeutic efforts to target SRC must consider the potential dire consequence of unintentionally also targeting YES, as the two cytoplasmic kinases have distinct and even opposing roles in the vasculature.

Results
YES regulates shear stress mediated phosphorylation of VE-cadherin Tyrosine phosphorylation of VE-cadherin is high in venules but low in arteries, correlating with the different ow patterns in these vessel types 17 . We investigated the pattern of VE-cadherin phosphorylation (pVE-cadherin) at Y685 in an intact vascular network (whole-mounted developing mouse retina at postnatal day 6; P6), in relation to the relative shear stress level. This was done using computational ow simulation on the PolNet platform 30 (Figure 1a, b), which models ow based on the vascular plexus geometry and the rheological properties of blood. The speci city of the pY685 antibody and other pVE-cadherin antibodies used in this study was validated in VE-cadherin mutant mice in which the corresponding tyrosines were mutated to phenylalanine (Supplementary gure 1a-c). The retinal vasculature was segmented into 5 regions: sprouting front (region 1); vein (region 2); capillaries near vein (region 3); capillaries near artery (region 4); artery (region 5) (Figure 1b). Shear stress was the lowest at the sprouting front, correlating to a very low pVE-cadherin signal. In veins and capillaries (regions 2-4), with low to medium shear stress levels, the pVE-cadherin signal was markedly induced. In the artery, with very high shear stress, the pVE-cadherin signal returned to low levels ( Figure 1b). In agreement, exposure of human umbilical vein endothelial cells (HUVECs) to low shear stress (3 dyn/cm 2 ) induced accumulation of pVE-cadherin, whereas at high shear stress (20 dyn/cm 2 ), VE-cadherin was unphosphorylated (Figure 1c, d). Furthermore, the appearance of pVE-cadherin in HUVECs correlated with phosphorylation of Src family kinases (SFKs) at Y418, indicative of SFK activity, which was induced under low shear stress and suppressed under high shear stress (Figure 1e-g). A direct and speci c relationship between SRC activation and phosphorylation of VE-cadherin has been suggested 16 , however the amino acid sequence around Y418 in the kinase domain of SRC is identical in multiple SFKs. YES, another SFK which is expressed by ECs, was preferentially localized to EC junctions while SRC showed a broad cytoplasmic and nuclear distribution ( Figure 1h). Moreover, YES transcript levels were higher than SRC in HUVECs (7.8±0.5 fold; Figure 1i). Expression of both SRC and YES was further upregulated by low shear stress ( Figure 1e). To compare the roles of YES and SRC in regulation of VE-cadherin phosphorylation, YES or SRC silenced HUVECs were subjected to 3 dyn/cm 2 shear stress. The silencing of either YES or SRC did not affect the expression of the other (Supplementary gure 1d, e). However, YES silencing strongly suppressed (82%) ow-induced pY685 phosphorylation, while SRC silencing resulted in a partial loss (58%) of pY685 levels (Figure 1j, k).

YES controls in vivo phosphorylation of VE-cadherin on Y658, Y685, and Y731
To explore the role of YES in vascular function in vivo, a conditional Yes knockout mouse line (Supplementary gure 2a) was crossed with Cdh5CreER T2 mice 31 , to allow tamoxifen-induced EC-speci c elimination of Yes expression (denoted Yes iECKO; see Supplementary gure 2b for knockout e ciency). Phosphorylation of VE-cadherin at the different sites, Y658, Y685 and Y731, was investigated by immunostaining. VE-cadherin phosphorylation in Yes iECKO retinas from P6 mice was down-regulated by 74±15%, 78±10% and 72±23% at Y658, Y685 and Y731 respectively, compared to littermate controls (Figure 2a-f; see Supplementary gure 1a-c for pY685 and pY731 antibody speci city; the pY658 antibody has been described 17 ). EC-speci c Src knockout (Src iECKO) affected VE-cadherin phosphorylation in P6 retinas to a lesser extent; pVE-cadherin levels decreased by 49±15% (Y658), 39±6% (Y685) and 46±7% (Y731) (Figure 2g-i, Supplementary gure 2c-e). In the Yes iECKO mice, pVE-cadherin levels were dramatically reduced also in large veins such as the vena cava (Supplementary gure 2f, g). Moreover, losing only one allele of Yes in ECs led to about 60% decrease in pY658 and pY685 VE-cadherin levels in P6 Yes wt/ : Cre+ retinas (Supplementary gure 2h, i). The long-term effect of endothelial Yes de ciency was studied in P22 mice which had been treated with tamoxifen at P1-3. The decrease in VE-cadherin phosphorylation was still evident 3 weeks after the recombination (Supplementary gure 2j, k), which shows that other SFKs did not compensate for Yes-de ciency during this time period.
To study the cell autonomous function of Yes, we introduced the Rosa26R-YFP reporter into the Yes iECKO mouse line. One dose of tamoxifen at P3 generated chimeric recombination, labelling most, but not all, of the Yes-de cient ECs with YFP. In the retinal vasculature of these mice, YFP+ ECs showed 45±12% lower pY685-VE-cadherin levels than the adjacent YFP-ECs (Figure 2j, k), indicating that YES controls VE-cadherin phosphorylation in a cell-autonomous manner.
Next, an inducible " oxed-STOP" Yes overexpression mouse model (denoted Yes iECOE) was generated by insertion of the complete mouse Yes cDNA with an upstream oxed transcriptional stop signal into the Hipp11 (H11) locus under the control of the CAG promoter (Supplementary gure 3a). By crossing with the Cdh5CreER T2 mouse and treatment with tamoxifen, EC-speci c Yes overexpression was achieved. The overexpressed YES protein localized to endothelial junctions in the retinas of Yes iECOE mice (Supplementary gure 3b), accompanied by a strong increase in the pY685 VE-cadherin signal. Strikingly, the pY685 VE-cadherin signal was markedly increased at the sprouting front as well as in arteries in Yes iECOE P6 retinas, to a level that was comparable to that in veins and capillaries (Figure 2l, m). In mice with chimeric induction of Yes overexpression, ECs displayed a several-fold higher level of VE-cadherin phosphorylation even in capillaries and veins where pVE-cadherin was already high in the control ( Figure  2n, o). This indicates that in WT mice, the extent of VE-cadherin phosphorylation relative to the total VEcadherin pool is limited. Thus, assuming that VE-cadherin in the Yes iECOE mouse is fully phosphorylated, less than 35% of total VE-cadherin pool was phosphorylated in veins and capillaries in the WT. Therefore we conclude that the limiting factor in ow-regulated VE-cadherin phosphorylation is the expression levels of YES. Interestingly, in the chimeric Yes iECOE mouse retina, branch points on both veins and arteries were more likely to be occupied by Yes OE cells (Supplementary gure 3c, d), indicating that ECs with higher levels of pVE-cadherin may be more resilient to disturbed ow shear stress at vessel bifurcations (note WSS changes at vessel branch points in Figure 1a, rightmost panel). Combined, these results suggest that YES is the main kinase regulating VE-cadherin phosphorylation in vivo.

YES-dependent VE-cadherin phosphorylation is dispensable for retinal vascular development and angiogenic sprouting
To study the impact of YES-dependent VE-cadherin phosphorylation on vascular development, we compared the developing retinal vasculature in different mouse models with loss-or gain-of-function of YES or VE-cadherin. Endothelial VE-cadherin null mice (Cdh5 iECKO) exhibited a strong hyper-sprouting phenotype (Figure 3a, b). In mice carrying a VE-cadherin mutation at a single phosphosite (Y685F), the postnatal retinal vessel development was unaffected (Figure 3c, d). In Yes iECKO mice, where phosphorylation of VE-cadherin was markedly reduced, the vascular development was normal compared to Cre+ control mice (Yes wt/wt , Cdh5CreER T2 +) (Figure 3e, f). In agreement, VEGFA induced angiogenic sprouting from aortic rings from both control and Yes iECKO mice (Supplementary gure 4a, b), but Yes iECKO sprouts were longer (Supplementary gure 4c). In contrast to the intact vascular development in the Yes iECKO retina, the retinal vascular area was signi cantly reduced in Yes iECOE mice, compared to control (H11-Yes 0/0, Cdh5CreER T2 +) (Figure 3g, h). We conclude that VE-cadherin is strictly required but its phosphorylation is dispensable for normal sprouting angiogenesis. Moreover, excessive phosphorylation of VE-cadherin interferes with vascular development.

VE-cadherin dependent collective cell migration is impaired by Yes deletion
During vascular development, ECs migrate from veins towards arteries 32 . Furthermore, endothelial cells migrate collectively in a manner dependent on the plasticity of adherens junctions 33 . To explore whether YES-mediated VE-cadherin phosphorylation impacts endothelial vein to artery migration, the localization of Rosa26R-YFP labelled ECs in Yes iECKO retinas was determined after partial recombination induced at P3 (Figure 4a). To localize YFP+ ECs in at-mounted retinas, arteries and veins were segmented (Supplementary Figure 5a). The distribution of YFP pixel density between veins and arteries was then mapped in a 2D coordinate system where veins were assigned a referential of 0 and arteries of 1. In this coordinate system, a referential of 0.5 marks a position at an equal distance from both veins and arteries.
In control P7 retinas, the YFP+ ECs were equally distributed between veins and arteries, assuming a reference point of 0.45 (Figure 4a, b, d), whereas in Yes iECKO retinas, the highest YFP+ EC density was found in arteries and there was a general shift of YFP signal towards the arterial side (Figure 4a, c, d; note loss of YFP+ ECs in veins in the Yes iECKO retina in Figure 4a, middle panel). The arterial-favored distribution was not present when recombination was induced at P6, 24 hours before retinas were collected ( Figure 4d; Supplementary gure 5b-d). Thus, although vascular development was morphologically unaffected, Yes-de cient ECs were privileged to migrate faster from veins to arteries in the Yes iECKO retina. In retinas with chimeric deletion of VE-cadherin, certain arteries were entirely devoid of VE-cadherin expressing ECs (Supplementary gure 5e), in agreement with that that loss of pVEcadherin-dependent cell-cell adhesion promoted migration towards arteries.
To determine if Yes-de ciency affects EC migration in vitro, HUVECs transfected with control or YES siRNA were analyzed in a monolayer scratch assay. Individual cells from the rst three rows from the migratory front were tracked and grouped by their initial position (Figure 4e; Supplementary video 1). Control ECs in the different rows migrated collectively at a similar speed while by silencing Yes, the cells in the rst and second rows separated from neighbouring cells and migrated at a faster speed ( Figure 4f). Thereby, YES-silenced cells from the rst and second rows migrated further than the control cells ( Figure  4g).
We further investigated whether the lack of VE-cadherin phosphorylation would affect other cellular processes involved in angiogenesis. EC proliferation in developing retinas was assessed by EdU incorporation followed by immunostaining of endothelial speci c nuclear protein ERG. The fraction of proliferative ECs was not changed in Yes iECKO mice compared to control (Yes / , Cdh5CreER T2 -) (Supplementary gure 5f, g). Moreover, deleting Yes in ECs did not affect the number of apoptotic cells in postnatal retinas, as demonstrated by immunostaining of cleaved caspase3 (Supplementary Figure 5h, i). Combined, these data show that Yes-de ciency impacts collective EC migration in vivo and in vitro while proliferation and apoptosis remain unaffected.

YES-dependent effects on adherens junction morphology
Endothelial cell adaptation to microenvironmental demands for remodelling, perfusion or permeability, is dependent on dynamic changes at adherens junctions, which appear as linear, jagged or JAIL (junction associated intermittent lamellipodia) junction morphologies 34  YES, but not SRC, is required for VE-cadherin internalization Next, the dynamics of VE-cadherin movements in the presence and absence of YES was studied by livecell microscopy. GFP-tagged VE-cadherin was transduced by lentivirus into HUVECs followed by siRNA silencing of YES and live-cell imaging. In control cells, VE-cadherin was detected in junction-proximal endocytic vesicles, which moved inwards from the junction. The internalized VE-cadherin vesicles disappeared within 2-5 minutes ( VE-cadherin internalization from the cell surface was further followed by employing an antibody feeding assay. Constitutive internalization of VE-cadherin was e ciently inhibited by silencing YES in HUVECs, whereas silencing of SRC left VE-cadherin unaffected (Figure 6b, c). Also, VEGFA-induced VE-cadherin internalization was blocked by YES-silencing (Figure 6d; supplementary gure 7a). VE-cadherin endocytic vesicles enriched in pVE-cadherin were moreover detected in vivo, just below the plasma membrane, in vena cava ECs from control mice. The frequency of endocytotic vesicles was markedly reduced in Yes iECKO mice (Figure 6e, f). Combined, these results indicate that phosphorylation of VE-cadherin is strictly required for its constitutive internalization, which is a prerequisite for junctional plasticity. p120 catenin has been implicated in regulation of VE-cadherin endocytosis. The degree of p120 catenin complex formation with VE-cadherin was assessed by co-immunoprecipitation between the two, which showed similar levels of complexes in the basal condition and a trend towards increased p120-VEcadherin association in response to VEGFA in both siControl and siYES HUVECS (Supplementary Figure  7b, c). Thus, complex formation with p120 was not affected by loss of VE-cadherin tyrosine phosphorylation.

Structural analyses of Yes-de cient endothelial junctions
The decreased VE-cadherin internalization in Yes-de cient ECs could potentially impact junction organization, which was addressed by transmission electron microscopy (TEM) analysis on the mouse ear dermis subjected or not to intradermal injection of VEGFA. Endothelial junction length was not affected by Yes deletion or VEGFA treatment (Figure 7a

Vascular permeability is increased in Yes iECKO mice
Phosphorylation of VE-cadherin has been implicated in vascular permeability and by studying individual, constitutively expressed VE-cadherin tyrosine to phenylalanine mutants the different phosphosites have been implicated in various regulatory aspects 10, 19-21 . As VE-cadherin is devoid of phosphotyrosine in the Yes iECKO mice, we could now assay the effects of VE-cadherin phosphorylation on the vascular barrier function in the context of the unphosphorylated, wildtype protein. The basal level of vascular permeability over time was tested by tail vein injection of Evans blue followed by a 2-hour circulation period. Higher levels of Evans blue in the interstitial tissue of the ear skin were detected in Yes iECKO mice (Supplementary gure 9a). The blood-brain-barrier integrity, assessed by the leakage of both Evans blue and Cadaverine into the cerebral tissue, was unaffected by Yes de ciency (Supplementary gure 9b, c). Thus, the vascular barrier was compromised in the YES-de cient dermis and potentially other tissues, but not in the CNS.
The permeability of developing retinal vessels was studied by cardiac injection of xable uorescent tracer (Dextran-TRITC, 10 kDa) in P6 mice. Leakage of the tracer into the perivascular tissue was observed preferentially at the vascular sprouting front and neither Yes deletion nor overexpression changed the number of leakage sites at the sprouting front (Figure 8a, b). However, an increased frequency of venous leakage was observed in Yes iECKO retinas (8 leakage sites observed in 11 veins from 10 retinas) in comparison to the control retinas (2 leakage sites observed in 14 veins from 9 retinas) (Figure 8a, c). In contrast, leakage from veins was not elevated in Yes iECOE retinas ( Figure 8c).
Next, permeability in mature vessels was investigated in adult mice by intravital confocal imaging of dermal vessels. In the wildtype dermis, pVE-cadherin was detected in some but not all venous vessels an also, unexpectedly, in some arterioles. This is in contrast to the pattern observed in the retina, where veins but not arteries display pVE-cadherin signals. The detection of pVE-cadherin in dermal arterioles may re ect ow-patterns that do not conform to the expected high ow-low phosphorylation pattern, but this remains to be determined. Still, VE-cadherin phosphorylation in dermal vessels was markedly reduced in adult Yes iECKO mice (Supplementary gure 9d, e), similar to the ndings in the developing retina. To monitor the leakage, a uorescent tracer was injected into the tail vein and leakage from dermal vessels in the ear was monitored after local injection of VEGFA (Figure 8d, Supplementary video 8, 9). Leakage was initiated at individual sites in venules and capillaries in both control and Yes iECKO mice (Figure 8e, Supplementary video 10, 11). The number of leakage sites was increased in both venules and capillaries in Yes iECKO mice compared to the control (Figure 8f, g). The uorescent intensity pro le at leakage sites showed that leakage appeared sooner after VEGFA injection in Yes iECKO mice compared to the control and that leakage was more vigorous in the absence of Yes (Figure 8h). The number of leakage sites was unaffected in Yes iECOE mice compared to wildtype (Figure 8i, j), indicating that leakage is not directly regulated by VE-cadherin tyrosine phosphorylation. In contrast to the increased leakage observed in Yesde ciency, Src iECKO mice exhibited a reduced number of leakage site after VEGFA injection compared to control (Figure 8k), in agreement with the presence of non-overlapping downstream targets for SRC and YES and that molecules other than VE-cadherin are involved in maintaining vascular barrier.
The Y731 phosphosite in VE-cadherin is required for leukocyte extravasation. Leukocytes populations were analysed by FACS in the uids collected by peritoneal lavage in mice 24 h after saline or VEGFA had been injected. In agreement with the phenotype of Y731F mice 10 , monocyte (CD11b+ LY6G-) extravasation was reduced in VEGFA treated Yes iECKO mice compared to control (Figure 8l).
Combined, these result show that loss of Yes-mediated phosphorylation of VE-cadherin results in dysregulation of endothelial junctions with increased leakage of macromolecules but suppressed extravasation of in ammatory cells.

Discussion
Here, we show that endothelial Yes-de ciency leaves VE-cadherin unphosphorylated at all known phosphotyrosine sites under conditions such as intermediate shear stress and agonist treatment that induce phosphorylation in the wildtype context. As a result, VE-cadherin endocytosis is blocked. The resulting junctional inertia interferes with endothelial collective migration during development, and disturbs barrier function. Basal as well as induced leakage of macromolecules is elevated in Yes-de cient ECs and occurs with abnormal kinetics, while leukocyte extravasation is suppressed. In contrast, overexpression of YES, leading to increased phosphorylation of VE-cadherin in all vessel types, does not affect vascular leakage. Therefore, YES-mediated tyrosine phosphorylation is critical for VE-cadherin's constitutive internalization and junctional pliability, but it is not required for formation of junctional gaps and vascular leakage.
Regulation of endothelial junctions has been ascribed to SRC 35,36 . The viral form, v-SRC, localizes to adherens junctions 37 , and gap junction-dependent cell-cell communication is regulated by c-SRC 38 .
Although broad screens for SRC kinase substrates reveal hundreds of molecules yet to be functionally characterized 39 , only a handful have become validated over several decades, including p85-Cortactin, p110-AFAP1, p130Cas, p125FAK and p120-catenin 27 . A common denominator for several of these SRC substrates is the involvement in actin-dependent processes, often with consequence for cell-matrix adhesion. In agreement, EC-speci c loss of Src expression results in impaired adhesion properties 26 .
Even less is known about YES substrates. YES phosphorylates the tight junction component occludin both in epithelial and endothelial cells 40 , but the functional consequence is unclear. YES has also been implicated in negative regulation of the cell cycle by phosphorylating cyclin dependent kinase 4 41 . As inferred from in vitro analyses, YES plays a role in T-lymphocyte immunity by phosphorylating CD46 42 and Collapsin Response Mediator Protein 2 (CRMP2) 43 . However, using in vitro kinase-based screen to identify substrates for SRC family kinases, no speci c substrate for YES was identi ed 44 . The important role for YES in ECs is suggested by that YES is expressed at considerably higher levels in HUVECs compared to SRC; single-cell transcriptomics shows that Yes is more abundant in ECs than in other cell types in a range of mouse tissues (https://tabula-muris.ds.czbiohub.org/https://tabulamuris.ds.czbiohub.org/) 45 .
An important conclusion from the results presented here is that the different SFKs must be targeted individually, e.g. in a therapeutic context but also in scienti c studies, as they serve different, sometimes opposing purposes. Yes-de ciency resulted in loss of VE-cadherin phosphorylation, suppressed endocytosis, clustering of VE-cadherin and increased macromolecular leakage. In contrast, in Srcde ciency, VE-cadherin phosphorylation was reduced albeit to a lesser extent, but VE-cadherin internalization remained unaffected and macromolecular leakage was suppressed. Combined, these data challenge the cause-consequence relationship between VE-cadherin internalization and formation of junctional gaps and macromolecular leakage, which has been the prevailing consensus. Speci cally, the loss of the Y685 site has been shown to prevent tracer leakage in healthy and pathological tissues 10,21,46 . Furthermore, intra-vital studies of mouse dermal vessels show that VEGFA or histamine-induced acute vascular leakage in mature vessels is established at predetermined sites in capillaries and post-capillary venules, where pY685 levels are high 5,47 . In contrast, in the developing retina vasculature, constitutive leakage occurs only at the sprouting front where pY685 level is low. Interestingly, in Yes iECKO mice with non-phosphorylated VE-cadherin, leakage was established also at retina venules. On the other hand, VEGFA-induced leakage was not increased in Yes iECOE mice with the increased pVE-cadherin pool. Combined, these results indicate that VE-cadherin phosphorylation is required for constitutive endocytosis of VE-cadherin and that it sensitizes ECs for rapid VEGFA-induced endocytosis and gap formation at EC junctions. Still, an additional trigger is required for gap formation and macromolecular leakage. Importantly, phosphorylation of VE-cadherin could also be a prerequisite for gap closure. In conclusion, the dysregulated leakage from Yes-de cient vessels may be a consequence of uncoordinated opening and closure of junctional gaps in addition to incomplete sealing in the resting state, due to loss of plasticity, akin to a stiff rubber band that has lost its normal elasticity.
The mechanism underlying the distinct effects of YES and SRC on VE-cadherin may be related to their different subcellular localizations, allowing phosphorylation of different subcellular pools of VE-cadherin, in addition to other kinase-speci c substrates. Thus, SRC was broadly distributed in the cytoplasm, while YES accumulated at junctions in HUVEC cultures. SRC is also localized to focal adhesions in freshly seeded ECs where it phosphorylates p125FAK 27 . Endothelial FAK-de ciency results in dismantling of adherens junctions 48 , and VEGFA-induced vascular leakage requires FAK induced phosphorylation of βcatenin 49 , underscoring the important role of matrix adhesion in regulation of EC junctions. The subcellular localization of cytoplasmic tyrosine kinases is in uenced by NH2-terminal fatty acid modi cation. YES is palmitoylated, while SRC is myristoylated, of consequence for different intracellular tra cking 50,51 . The different subcellular localizations of YES and SRC would also be compatible with their distinct regulation of the actin cytoskeleton as shown here: Yesbut not Src-de cient HUVECs exhibited an increase in cortical actin and loss in actin stress bers in response to VEGFA. In agreement, VEGFA treatment increased the electron dense, cortical actin-enriched junctional area in Yes-de cient dermal vessels. The abnormal actin arrangements in the absence of Yes, provides a potential mechanistic underpinning for the changes in junctional dynamics (reduced jaggedness and increased JAIL formation) observed in vivo and in vitro, and indeed, the block in VE-cadherin internalization.
Binding of p120-catenin to a VE-cadherin juxtamembrane motif has been implicated in VE-cadherin stabilization 13 , and in agreement, dephosphorylation of VE-cadherin at Y658 increases the binding a nity for p120 in vitro 18,52 . Moreover, the vascular barrier is normal in mice with a mutation of this endocytic motif in VE-cadherin but the vessel density is decreased in the developing retina 53 . In agreement with previous work on phosphorylation-de cient VE-cadherin models 17 , we failed to detect changes in the binding of p120 to VE-cadherin in the absence of YES, in which constitutive VE-cadherin internalization was markedly inhibited. Importantly, in contrast to the VE-cadherin endocytosis de cient mutants described previously 53 , the expression levels of both VE-cadherin and p120 were unaffected by Yes-deletion.
The exaggerated macromolecular leakage in Yes -/vessels is at odds with the leakage phenotype of Y685F mice which show suppressed leakage of macromolecules 10,21 and suppressed extravasation of neutrophils 20 . The difference in phenotypes between the Y685F and the Yes-de cient mouse models may be related to the fact that Yes-de ciency affects additional molecular regulators at endothelial junctions, yet to be discovered. However, the block in monocyte extravasation is in agreement with the phenotype of the Y731 vasculature 10  Intravital vascular leakage assay. Intravital imaging of the mouse ear dermis with intradermal injection has been described previously 5 . Brie y, following systemic administration of 2000-kDa Dextran by tailvein injection, mice were anaesthetised by intraperitoneal injection of Ketamine-Xylazine (120 mg/kg Ketamine, 10 mg/kg Xylazine) to a surgical level and the ear secured to a solid support. Mice were maintained at a body temperature of 37°C for the entire experiment, maximum 90 minutes. Time-lapse imaging was performed using single-photon microscopy (Zeiss LSM 710) and a high N.A waterimmersion lens (CF175 apochromat 25xW N.A1.1, Nikon). For intradermal EC stimulation, a volume of approximately 0.1 μl recombinant mouse VEGFA164 (Peprotech), concentration 100 ng/μl, was injected using a sub-micrometer capillary needle. 10 kDa TRITC Dextran was used as a tracer for VEGFA. Leakage sites were identi ed in time-lapse imaging, following VEGFA injection, as de ned sites of concentrated dextran in the extravascular space. The researcher was blinded to the genotype of the mice when performing the analyses.
In vivo wall shear stress analysis. Flow simulation from the images of retinal vasculature shown by CD31 staining was done using the PolNet platform 30 . Brie y, at mounted retinas were imaged at an area including one artery, one or two veins and the complete capillary network in between by tile scanning confocal microscope (Leica SP8) using 40X objective. 3D reconstruction of the vascular network was done using maximum projection of the vessels illustrated by CD31 immunostaining. A heat map of relative shear stress level was generated based on the ow simulation by assigning inlet at the artery and outlet at the veins. Regional average shear stress levels were obtained and plotted combined with the pVE-Cadherin levels.
Retinal EC distribution analysis. Chimeric recombination was induced at P3 or P6 by i.p. injection of tamoxifen (100 µg/mouse, Sigma). Retinas were taken at P7 and stained for CD31 and YFP and then at mounted to slides by cutting into 4 lea ets. Images of whole retinas were taken by z-stack tile scanning using 10X objective on a confocal microscope (Leica SP8). Maximum projection images of whole retinas were used for image segmentation, which was performed with resource to ImageJ. Maximum projection of YFP channel threshold was established to distinguish YFP+ cells from the background signal. Outliers with radius between 0.2-1 µm were removed. CD31 channel (after maximum projection) was used to de ne the outline of the retina, veins and arteries, and the optic nerve as a mask to de ne a referential system. For computational analysis a bespoken python-based work ow 2D-coordinate system was employed, that can be accessed on github (https://github.com/wgiese/EC_migration_retina). For every pixel in the microscopic image, three numbers were computed (using the mask as referential): (1) distance to the nearest vein (dVD), (2) distance to the nearest artery (dAD) and (3) radial distance to the optic nerve (r). From these measures, the relative distances by dAV = dVD/(dVD + dAD), were obtained. The EC distribution was computed by performing the operation for all YFP positive pixels, which were used as a proxy for EC distribution.
Vessel leakage in postnatal retina. Recombination was induced at P1-3 by i.p. injection of tamoxifen (100 µg/day/mouse, Sigma). At P6, mice were anesthetized and 10 kDa TRITC-Dextran-Lysine (100 µg/g body weight, TdB Labs) was injected into the left heart ventricle. After 10 minutes of circulation, mice were decapitated and eyes were taken and xed in 4% paraformaldehyde (PFA) (Sigma) for 2 hours at room temperature. A needle puncture on the cornea was made to facilitate fast xation of the retinas. Retinas were stained with CD31 to visualize blood vessels. Images were acquired by z-stack tile scanning in a confocal microscope (Leica SP8). A site of leakage was scored positive if dextran was detected in the perivascular region.
Immune cell in ltration assay. Recombination was induced in 8-12 weeks old mice by gavage of tamoxifen (2 mg/day/mouse, Sigma) for 5 consecutive days. The mice were allowed rest for 2 days and saline or VEGFA (5 µg/kg body weight) was administered through i.p. injection. After 24 h, mice were sacri ced by cervical dislocation and ve mL of phosphate buffered saline (PBS) was subsequently injected intraperitoneally, and the suspension of recruited cells was collected after abdominal palpation.
Transmission Electron microscopy. 8-12 weeks old control or Yes iECKO mice were treated with intradermal injection of VEGFA164 (10 ng, Peprotech) in the ear skin. The other ear served as an untreated control. Mice were then anesthetized and perfused rst with 10 mL Hank's balanced salt solution (HBSS) and then 12 mL cold xative (1% glutaraldehyde (Sigma), 4% PFA in 0.1 M phosphate buffer) through the left ventricle. Ears were cut and placed in the xative for 30 min at 4°C and then washed 3 times with PBS, dehydrated and embedded in Epon. Imaging was done with Tecnai G2 Spirit 120 kV transmission electron microscope and Quamesa CCD camera.
Aortic ring assay. Thoracic aortas were taken from mice at P7 and cut into 1 mm rings after removal of the connective tissues. The rings were embedded between two layers of rat tail collagen, type I (ThermoFisher Scienti c) and cultivated in Dulbecco's Modi ed Eagle's Medium (DMEM) (Gibco) supplemented with 10% foetal bovine serum (FBS) and VEGFA (Peprotech, 30 ng/ml). 4-OH-tamoxifen (5 µg/mL, Sigma) was added to the culture for 48 hours to induce recombination. After 4 days of initial cultivation samples were xed in 4% PFA (Sigma) and analysed by microscopy after immunostaining with antibodies.
Proliferation. Proliferation in the developing retinas was assessed by i.p. injection of 5-ethynyl-2′deoxyuridine (EdU, 100 µg/mouse, ThermoFisher Scienti c) at postnatal day 6. Mice were sacri ced 2 hours after injection and retinas were removed and xed in 4% PFA (Sigma). EdU staining on retinas was done according to the instructions for the Click-iT EdU imaging kit (ThermoFisher Scienti c) apart from 3 h incubation in the reaction mix at room temperature. The samples were co-stained with antibodies to ERG to identify EC nuclei. Images were taken by confocal microscope (Leica SP8) and analyzed by ImageJ.
Lungs were taken at P6 and dissociated to get cell suspension using the MACS dissociation kit (Miltenyi Biotec). Isolation of ECs from mouse lung was done using anti-CD31-antibody (BD Pharmingen)conjugated dynabeads (ThermoFisher Scienti c). RNA from the isolated ECs was extracted immediately by using RNeasy mini plus kit (QIAGEN).
Cell culture. HUVECs (PromoCell) were cultured at 37°C with 5% CO 2 in endothelial cell growth MV2 with supplied supplements (PromoCell). Cells were tested for cell morphology and cell-type speci c markers using ow cytometric analyses by the vendor. For VEGFA stimulation, con uent cells were starved in MV2 with 0.2% FBS without supplements for 5 hours. The same medium with VEGFA165 (50 ng/ml, Peprotech) was then added to the cells and incubated at 37°C for indicated time periods.
Scratch wound healing assay. HUVECs were seeded at a density of 50,000 cells/well into a 96-well ImageLockTM tissue culture plate (Essen BioScience) and incubated in MV2 medium overnight followed by standardized scratching using WoundMakerTM (Essen BioScience). Cells were washed with PBS and the plate was placed into an IncuCyte ZOOM (Essen BioScience) and scanned every 15 min for 12 h using a 10× objective during which data were collected using ZOOM software (Essen BioScience).
Cellular migration was analyzed using MTrackJ in ImageJ by manually marking the cells at each time frame. Migration speed of one cell was obtained by dividing the migration distance by time.
Confocal live cell imaging. HUVECs were seeded at full con uency on glass-bottom plates (MatTek) and cultured in MV2 medium overnight. Fresh medium with SiR-actin dye (Cytoskeleton) was added to the cells and incubated for two hours. The cell monolayer was scratched using a 2 µL pipette tip and changed to fresh medium. The plates were then placed on a Leica SP8 microscope equipped with a humidi ed CO 2 incubator (Leica Microsystem). Time-laps z-stack images were scanned every 1 min for 3 hours.
In vitro ow treatment. HUVECs were seeded at full con uency on µ-slides (Ibidi) coated with bronectin (1 µg/mL, Sigma) and cultured overnight in degassed MV2 medium. The slides were then connected to the Ibidi pump system to allow degassed medium owing through at 3 or 20 dyn/cm 2 shear stress for 24 h at 37°C in a CO 2 incubator. Static control cells were cultured on the slides in the incubator for the same time period. Cells were xed in 1% PFA for 10 min for immuno uorescence or lysed in RIPA buffer (ThermoFisher Scienti c) for western blots.
VE-cadherin internalization assay. Control or YES silenced con uent HUVECs were incubated with an antibody against the VE-cadherin extracellular domain (Clone BV6, MABT134, Merck Millipore, 10 µg/mL) at 4°C for 1 hour in MV2 basal medium containing 3% bovine serum albumin (BSA) (Sigma) without supplements. Unbound antibody was removed by rinsing cells in ice-cold MV2 basal medium. Cells were then cultured at 37°C for 4 hours in the presence of 150 µM chloroquine (Sigma) in MV2 medium with supplements followed by washing for 8 min with PBS containing 25 mM glycine and 3% BSA with Ca 2+ , Mg 2+ , pH 2.7, to remove the antibody from cell surface. Cells were rinsed with PBS and xed with 2% PFA for 10 min at room temperature. Total VE-cadherin was stained with an antibody generated in another species (AF1002, R&D Systems). Samples were mounted on the slides with Fluoromount G (Southern Biotech) and analyzed by confocal imaging.
RNA interference. siRNAs targeting human YES or SRC (Sigma) were transfected into HUVECs by using Lipofectamine RNAiMAX (ThermoFisher Scienti c). Control cells were transfected with control siRNA which does not target any speci c gene. siRNAs were mixed with lipofectamine added to the cells at a concentration of 10 nM and incubated overnight before changing to fresh medium. The cells were used for experiments 48 h after transfection. 80-95% knockdown e ciency was validated by quantitative-PCR. YES siRNA: SASI_Hs01_00086922; SRC siRNA: SASI_Hs01_00112907; Control siRNA: MISSION siRNA universal negative control (SCI001).
Immuno uorescence. Mice were sacri ced and tissues were immediately removed and xed in 2% PFA (eyes and ears were xed for 1 h, vena cava was xed for 15 min). Samples were incubated in PBS with 0.5% Triton X-100 and 1% BSA for 3 hours at room temperature and then with primary antibodies in the same solution overnight at 4°C with agitation. After 3 washes at room temperature, samples were incubated with secondary antibodies overnight at 4°C, washed 3 times and mounted to slides for imaging. For staining of cultured cells, cells were xed in 1% PFA for 10 min after treatment. After xation, the cells were washed 3 times in PBS and permeabilized in PBS with 0.1% TritionX-100 for 5 minutes, followed by block in 1% BSA for 30 minutes and incubation with primary antibodies overnight at 4°C. Cells were washed 3 times in 0.1% TritonX-100 and incubated with secondary antibodies for 1 hour at room temperature. They were then washed 3 times and mounted to slides for imaging. Specimens were imaged by Leica SP8 (Leica Microsystems) confocal microscopes. Image processing and analysis was done using ImageJ software.
Quantitative PCR. RNA was extracted and puri ed using RNeasy Plus kit (Qiagen). RNA concentrations were measured by Nanodrop spectrophotometer (ThermoFisher Scienti c) and adjusted to equal concentration, followed by reverse transcription using SuperScript III (ThermoFisher Scienti c). Real-time quantitative PCRs were performed on Bio-Rad real-time PCR machine using Sso advanced qPCR reagent (Bio-Rad). Housekeeping gene ribosomal protein L19 (Rpl19) was used as internal control. The comparative Ct method was used to calculate fold differences.
Rho-GTPase activity assay. Control or YES silenced HUVECs were starved at 37°C in basal MV2 medium with 0.2% FBS for 3 hours, and then stimulated with VEGFA164 (50 ng/mL, Peprotech) for 15 minutes.
Activities of Rho GTPases Rac1, CDC42 and RhoA were detected using kit from Cytoskeleton. Brie y, cells were lysed and 300 µg of freshly prepared total protein were incubated with Rhotekin-RBD (RhoA) or PAK-PBD (CDC42, Rac1) beads for 1 hour on a rotator at 4°C. Beads were washed 5 times and boiled in Laemmli buffer with β-Mercaptoethanol. The samples were separated on an SDS-PAGE (4-12% gradient gel) (ThermoFisher Scienti c) and western blots were performed to detect Rac1, CDC42 and RhoA respectively with the antibodies supplied in the kit.
Immunoprecipitation. Control or YES silenced HUVECs were starved at 37°C in basal MV2 medium with 0.2% FBS for 3 hours, and then stimulated with VEGFA164 (50 ng/mL, Peprotech) for 15 minutes. Cells were immediately lysed on ice in lysis buffer (50 mM Tris-HCl pH 7.5, 5 mM EDTA, 150 mM NaCl, 0.5% NP-40) supplemented with protease and phosphatase inhibitor cocktails (Roche). Lysates with 300 µg of total protein were incubated with 1.5 µg VE-cadherin antibody (AF1002, R&D Systems) overnight at 4°C. Protein G Sepharose beads (Cytiva) was added to the mixture and incubated for 3 hours at 4°C. Beads were washed 5 times with lysis buffer and proteins were denatured and released from the beads by boiling in β-Mercaptoethanol-containing Laemmli buffer.
Statistics and Reproducibility. Statistical analysis was performed using GraphPad Prism software.
Statistical signi cance in comparison between two groups was determined by unpaired two-tailed Student's t-test. Two-way ANOVA was used to compare between groups in time course experiments.
Variances were similar between the groups compared. Differences were considered signi cant with a P < 0.05. For animal experiments, no statistical methods were used for predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiment and outcome assessment. Data availability. The data that support the nding of this study are available from the corresponding authors upon reasonable request. Figure 1 Shear stress regulated VE-cadherin phosphorylation is mediated by YES     EC localization is affected by VE-cadherin or Yes deletion due to disturbed collective cell migration (a) Chd5Cre-induced expression of YFP (green) in wt/wt, Cdh5CreER T2 +, and Yes / , Cdh5CreER T2 +, at P7 after tamoxifen (100 µg/mouse) injection at P3. CD31 (magenta) counter staining was performed to show all ECs. Veins and arteries are shown enlarged to the right with vessel outlines marked by yellow