Tyrosine-protein kinase Yes 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 diseases such as cancer and retinopathies. Inhibitors of Src cytoplasmic tyrosine kinase have been applied to suppress VE-cadherin tyrosine phosphorylation and prevent excessive leakage, edema and high interstitial pressure. Here we show that the Src-related Yes tyrosine kinase, rather than Src, is localized at endothelial cell (EC) junctions where it becomes activated in a flow-dependent manner. EC-specific Yes1 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. Overexpression of Yes1 causes ectopic V E-ca dh erin p ho sp ho ry la tion, while vascular leakage is unaffected. In contrast, in EC-specific Src deficiency, VE-cadherin internalization is maintained and leakage is suppressed. In conclusion, Yes-mediated phosphorylation regulates constitutive VE-cadherin turnover, thereby maintaining endothelial junction p l a st i c ity a n d v as cu lar integrity.
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 flexible, allowing remodeling of the embryonic and postnatal vascular beds and exchange of small molecules and fluid to maintain adult tissue homeostasis 2 . During acute inflammation and in a range of chronic diseases, the barrier is weakened, leading to extravasation of blood constituents, promoting inflammation and progression of pathologies such as retinopathies and cancer [1][2][3] . The maintenance of junctions and barrier function relies on junctional complexes consisting of cell-adhesion molecules formed between adjacent ECs. Tight junctions limit vascular leakage preferentially in the central nervous system (CNS) and in arteries and post-arterial capillaries in peripheral tissues [4][5][6] . Adherens junctions, formed by VE-cadherin and associated catenins, are crucial for regulating venous and capillary vascular permeability, leukocyte extravasation and collective cell migration [7][8][9][10][11][12][13] . VE-cadherin is rapidly internalized in Article https://doi.org/10.1038/s44161-022-00172-z plexus geometry and the rheological properties of blood. The retinal vasculature was segmented into five regions: sprouting front (region 1); vein (region 2); capillaries near vein (region 3); capillaries near artery (region 4); and artery (region 5) (Fig. 1b). Simulation showed that 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 (Fig. 1b), in accordance with findings in organs such as the diaphragm, the bladder and the yolk sac 18,32 .
A similar pattern was seen in vitro; exposure of human umbilical vein endothelial cells (HUVECs) to low shear stress (3 dyn cm −2 )-induced pY685 VE-cadherin levels, whereas at high shear stress (20 dyn cm −2 ), VE-cadherin was unphosphorylated (Fig. 1c,d). The appearance of pVE-cadherin in HUVECs correlated to 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 ( Fig. 1e-g). A direct relationship between Src activation and phosphorylation of VE-cadherin has been suggested 17 ; however, the amino acid sequence around Y418 in the kinase domain of Src is identical in multiple SFKs. Yes, another SFK expressed by ECs, was preferentially localized to EC junctions, whereas Src displayed a weak junctional signal and broad perinuclear and nuclear distribution (Fig. 1h). Moreover, YES1 transcript levels were higher than SRC levels in HUVECs (7.8 ± 0.5-fold; Fig. 1i). The expression of both Src and Yes was further upregulated by low shear stress (Fig. 1e).
To compare the roles of Yes and Src in the regulation of VE-cadherin phosphorylation, YES1-or SRC-silenced HUVECs were subjected to 3 dyn cm −2 shear stress. Silencing of either YES1 (siYES1) or SRC (siSRC) did not affect the expression of the other (Extended Data Fig. 1d,e); however, siYES1 strongly suppressed (82%) flow-induced phosphorylation of VE-cadherin at Y685 and Y658 (Fig. 1j,k and Extended Data Fig. 1f,g), whereas siSRC resulted in a partial loss (58%) of pY685 levels (Fig. 1j,k). Treatment of HUVECs exposed to 3 dyn cm −2 flow with siYES1 moreover suppressed flow induction of the pY418 SFK signal, whereas treatment with siSRC was without effect (Fig. 1l,m), indicating that Yes but not Src, phosphorylates VE-cadherin at low shear stress. Single-cell RNA-sequence analysis of P6 and P10 retinal ECs showed overall higher expression of Yes1 than Src in vivo; however, the expression of both was relatively uniform between non-proliferative vessel subtypes (Extended Data Fig. 1h,i and Supplementary Table 1), whereas expression of VE-cadherin trended toward higher levels in arteries (Extended Data Fig. 1j). Combined, these data suggest that the different levels of VE-cadherin phosphorylation in arteries and veins (Fig. 1a,b) are not caused by different expression levels of Yes1, but by different extents of flow-induced Yes kinase activity.

VE-cadherin phosphorylation on Y658, Y685 and Y731 by Yes
A conditional, Cdh5CreER T2 -controlled 33 Yes1 knockout mouse (Yes1 iECKO; Extended Data Fig. 2a) was generated to eliminate endothelial expression of Yes1. The deletion efficiency with tamoxifen treatment was 70% (Extended Data Fig. 2b). In the Yes1 iECKO P6 retina, phosphorylation of VE-cadherin was downregulated by 74 ± 15%, 78 ± 10% and 72 ± 23% at Y658, Y685 and Y731, respectively, compared to littermate controls (Fig. 2a-f and Extended Data Fig. 1a-c for pY685 and pY731 antibody specificity; the pY658 antibody has been described 18 ). In EC-specific Src knockout (Src iECKO; 80% deletion efficiency 28 ) pVE-cadherin levels decreased by 49 ± 15% (Y658), 39 ± 6% (Y685) and 46 ± 7% (Y731) in the P6 retina ( Fig. 2g-i and Extended Data Fig. 2c-e). In Yes1 iECKO mice, pVE-cadherin levels were dramatically reduced in large veins, such as the vena cava (Extended Data Fig. 2f,g). Moreover, losing only one allele of Yes1 in ECs led to 60% decrease in pY658 and pY685 VE-cadherin levels in P6 Yes wt/fl :Cre + retinas (Extended Data Fig. 2h,i). The long-term effect of endothelial Yes deficiency was examined in P22 mice, treated with tamoxifen at P1-3. The loss in VE-cadherin response to inflammatory 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 the regulation of its internalization, the mechanisms of gap formation and vascular leakage have remained poorly understood 14,15 . This is important, as exaggerated vascular leakage and the formation of edema are drivers of disease 1 .
Three main tyrosine phosphorylation sites have been identified in VE-cadherin, Y658, Y685 and Y731 (refs. 16,17 ). Both Y658 and Y685 are phosphorylated in veins, but not in arteries, as a consequence of flow-induced activation of Src cytoplasmic tyrosine kinases 18 . Phosphorylated (p)Y658 relieves tension across VE-cadherin bridges through binding of the polarity protein LGN (leucine-glycine-asparagine repeat), which displaces VE-cadherin-associated p120-catenin 19 . Phosphorylation of Y685 is induced by flow as well as by inflammatory cytokines and VEGFA 17,18,20 . Vascular leakage requires phosphorylation of 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 and in disease models 11,21,22 . Leakage is accompanied by VE-cadherin endocytosis and reduced levels of pVE-cadherin 18 . Y731 is constitutively phosphorylated and its dephosphorylation by Src homology phosphatase 2 (SHP2) is a prerequisite for leukocyte extravasation 11,23 . Several additional phosphotyrosine phosphatases (PTPs), such as VE-PTP and density enhanced phosphatase 1 (DEP1; also denoted CD148), have been implicated in the regulation of VE-cadherin phosphorylation [24][25][26] .
Src-mediated VE-cadherin phosphorylation on Y685 in response to VEGFA stimulation has been regarded as an important step in VEGFA-induced leakage 11,17,19 ; however, EC-specific Src deficiency interferes with cell-matrix adhesion rather than cell-cell adhesion 27,28 . In addition to Src, ECs express the highly related cytoplasmic tyrosine kinases Yes and Fyn, the roles of which 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 29 . Accordingly, constitutive global gene inactivation results in abolished VEGFA-induced vascular leakage in constitutive gene inactivated Src −/− or Yes1 −/− mice but not in Fyn −/− mice 30 .
Here, we demonstrate a critical role for Yes in the phosphorylation of VE-cadherin at all main phosphorylation sites, Y658, Y685 and Y731. Loss of VE-cadherin phosphorylation suppresses its endocytosis and interferes with cell-cell adhesion, leading to abnormal barrier properties with blunted collective migration, excessive macromolecular leakage and suppressed leukocyte extravasation. Forced phosphorylation of VE-cadherin by Yes1 overexpression does not affect agonist-induced leakage; thus, macromolecular leakage is not controlled solely by the extent of VE-cadherin phosphorylation. In contrast, in Src deficiency, VE-cadherin endocytosis is unaffected and vascular leakage is reduced. Therefore, Yes and Src differentially regulate VE-cadherin turnover, adherens junction stability and vascular leakage. We conclude that the two highly related cytoplasmic kinases have distinct and even opposing roles in the vasculature.

Yes regulates shear stress-induction of pVE-cadherin
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 wall shear stress (WSS) level. The specificity of the pY685 antibody was validated in VE-cadherin tyrosine to phenylalanine, Y685F, mutant mice (Extended Data Fig. 1a). A similar strategy was used to validate the specificity of another pVE-cadherin antibody against pY731 (Extended Data Fig. 1b,c). Shear stress in different regions of the retina vasculature was estimated using computational flow simulation on the PolNet platform 31 (Fig. 1a,b), which models flow based on the vascular Article https://doi.org/10.1038/s44161-022-00172-z phosphorylation seen at P6 was still evident 3 weeks after recombination (Extended Data Fig. 2j,k), thus, other SFKs did not compensate for Yes1 deficiency during this time period.
To study the cell-autonomous function of Yes1, the Rosa26R-YFP reporter was introduced into the Yes1 iECKO mouse line. One dose of tamoxifen at P3 generated chimeric recombination, labeling most, but not all, of the Yes1-deficient ECs with YFP. YFP + retinal ECs showed 45 ± 12% lower pY685 VE-cadherin levels than adjacent YFP-ECs (Fig. 2j,k), indicating that Yes controls VE-cadherin phosphorylation in a cell-autonomous manner. The relatively limited reduction on pY685 VE-cadherin levels was expected as expression of the YFP reporter does not completely correlate with the Yes1 gene deletion.
VE-cadherin is a substrate for the EC-specific phosphotyrosine phosphatase VE-PTP (PTPRB) 24,34 . Silencing of PTPRB in HUVECs markedly increased pY685 VE-cadherin levels (Extended Data Fig. 3a-c). Treatment of HUVECs with siYES1 suppressed pY685 levels also in PTPRB-deficient ECs, indicating that Yes acts directly on VE-cadherin, rather than indirectly, by modulating VE-PTP activity.
Next, an inducible 'floxed-STOP' Yes1 overexpression mouse (denoted Yes1 iECOE) was generated by insertion of the mouse Yes1 complementary DNA with an upstream floxed transcriptional stop signal into the Hipp11 (H11) locus, controlled by the CAG promoter (Extended Data Fig. 4a). Following crossing with the Cdh5CreER T2 mouse line and treatment with tamoxifen, EC-specific Yes1 overexpression was achieved (H11-STOP-Yes1 +/0 , Cdh5CreER T2+ ). The overexpressed Yes protein localized to endothelial junctions in the retinas of Yes1 iECOE mice (Extended Data Fig. 4b), accompanied by a strong increase in the pY685 VE-cadherin signal. Strikingly, the pY685 VE-cadherin levels were increased at the sprouting front as well as in arteries in Yes1 iECOE P6 retinas to levels comparable to those in veins and capillaries (Fig. 2l,m). In mice with chimeric induction of Yes1 overexpression, pVE-cadherin levels increased several-fold also in retinal capillaries and veins where levels were already high in the control (Fig. 2n,o). This indicates that in wildtype (WT) mice, pVE-cadherin levels relative to total VE-cadherin protein is limited. Thus, assuming that VE-cadherin in the Yes1 iECOE mouse is fully phosphorylated, less than 35% of the total VE-cadherin pool was phosphorylated in veins and capillaries in the WT. We conclude that the limiting factor in flow-regulated VE-cadherin phosphorylation is the expression level of Yes, which when sufficiently high, allows its activation and phosphorylation of VE-cadherin also under conditions of high shear stress such as in arteries. Notably, in the chimeric Yes1 iECOE mouse retina, branch points on both veins and arteries were more likely to be occupied by Yes1 OE cells (Extended Data Fig. 4c,d), suggesting that ECs with high pVE-cadherin may be more resilient to disturbed flow at vessel bifurcations (note wall shear stress (WSS) changes at vessel branch points in Fig. 1a, right). Alternatively, Yes1 OE cells may be blunted in their ability to move away from bifurcations.
To compare the effects of Yes and Src overexpression on pY685 VE-cadherin, Src and Yes were introduced in HUVECs via lentiviral transduction. Forced overexpression of Src and Yes promoted increased pY685 VE-cadherin levels to a similar extent (Extended Data Fig. 5a-e); however, in Src-overexpressing cells, both Src and pY685 VE-cadherin were detected in the cytoplasm, concentrated in the perinuclear region, whereas in Yes overexpressing cells, both Yes and pY685 were detected at broad folds at endothelial junctions (Extended Data Fig. 5a; compare pY685 VE-cadherin immunostaining in Yes-and Src-overexpressing cells). These results suggest that Yes is the main kinase regulating VE-cadherin phosphorylation at EC junctions.

Yes1 controls endothelial collective cell migration
The plasticity of endothelial cell-cell adhesions is known to impact collective cell migration 35 . During vascular development, ECs migrate from veins toward arteries 36 . To explore whether Yes-mediated VE-cadherin phosphorylation impacts endothelial vein-to-artery migration, the dual reporter-Cre, iSuRe-Cre, was exploited for EC tracing 37 . In the iSuRe-Cre model, MdTomato-reporter expression is genetically linked with deletion of the target gene. The localization of MbTomato-labeled ECs in control and Yes1 iECKO retinas was determined at P7 and P15 after partial recombination induced at P3 (Fig. 3a). Artery and vein regions were segmented and the localization of MbTomato + ECs was mapped in two-dimensions, shown with the relative distance between veins and arteries on the x axis (vein, 0.0; artery, 1.0) and the distance from optic nerve to sprouting front on the y axis ( Fig. 3b). At P7, the vein-to-artery distribution of MbTomato + ECs in Yes1 iECKO retinas was similar to the control (Fig. 3c-e); however, in the radial distribution, MbTomato + ECs in Yes1 iECKO retinas were positioned closer to the optic nerve, compared to control MbTomato + ECs (Fig. 3c,f,g). At P15, MbTomato + ECs in Yes1 iECKO retinas were concentrated to arteries with only scarce presence in veins and capillaries (Fig. 3c-e). Moreover, at P15, Yes1 iECKO MbTomato + ECs were localized closer to the optic nerve than in the control (Fig. 3c,f,g). With a higher degree of recombination following multiple tamoxifen injections at P1, 3, 6 and 9, the concentration of Yes-deficient ECs in arteries was striking (Fig. 3h,i), accompanied by widening of the artery diameter in the distal region, away from the optic nerve (Fig. 3i,j). Thus, compared to WT ECs, Yes-deficient ECs are privileged to migrate toward arteries in the developing retina. In retinas with chimeric deletion of VE-cadherin, certain arteries were entirely devoid of VE-cadherin-expressing ECs (Extended Data Fig. 6a), in agreement with the finding that loss of pVE-cadherin-dependent cell-cell adhesion promoted migration toward arteries 38 .
A monolayer scratch assay was employed to investigate how Yes deficiency affects EC migration in vitro. Individual cells from the first three rows from the migratory front were tracked and grouped by their initial position (Extended Data Fig. 6b and Supplementary Video 1). Control ECs in the different rows migrated collectively at a similar speed, whereas by silencing YES1, cells in rows 1-3 separated from neighboring cells and migrated at a faster speed (Extended Data Fig. 6c) and YES1-silenced cells from the first and second rows migrated further than the control cells (Extended Data Fig. 6d). Next, the activities of the well-known regulatory factors for migration, Rho GTPases CDC42, Rac1 and RhoA were examined (Extended Data Fig. 6e-h). The basal activity of CDC42 was increased in YES1-silenced cells. Basal RhoA activity showed the same trend, while Rac1 was unaffected, indicating that perturbed activity of small GTPases contributed to EC dysregulation in Yes1 deficiency 39,40 .
The loss of YES-dependent VE-cadherin phosphorylation and endothelial migration did, however, not interfere with sprouting angiogenesis. In Yes1 iECKO mice, retinal vascular outgrowth at P6 was normal compared to Cre + control mice (Yes1 wt/wt , Cdh5CreER T2+ ) (Extended Data Fig. 7a,b), and the number of filopodia on tip cells at the sprouting front was unaffected (Extended Data Fig. 7c,d).
Retinal vascular outgrowth was unaffected also in mice carrying a Y685F VE-cadherin mutation (Extended Data Fig. 7e,f). In contrast, VE-cadherin-null mice (Cdh5 iECKO) exhibited a strong hypersprouting phenotype paralleled by delayed outgrowth (Extended Data Fig. 7g,h). Thus, the lack of VE-cadherin tyrosine phosphorylation in the Yes1 iECKO mouse does not result in a Cdh5 null phenotype. The retinal vascular area was reduced also in Yes1 iECOE mice compared to control (H11-STOP-Yes1 0/0 , Cdh5CreER T2+ ) (Extended Data Fig. 7i,j). Therefore, VE-cadherin is strictly required, but its phosphorylation is dispensable for normal sprouting angiogenesis. Moreover, excessive phosphorylation of VE-cadherin is accompanied by disturbed vascular development.
We further investigated whether the lack of VE-cadherin phosphorylation would affect other cellular processes involved in angiogenesis. VEGFA induced angiogenic sprouting from aortic rings from both control and Yes1 iECKO mice (Extended Data Fig. 8a,b), but Yes1 iECKO sprouts were longer (Extended Data Fig. 8c), in accordance with enhanced cell motility. EC proliferation in developing retinas was assessed by 5-ethynyl-2-deoxyuridine (EdU) incorporation followed by immunostaining of the endothelial-specific nuclear protein ERG. The

Adherens junction morphology in Yes and Src deficiency
EC adaptation to microenvironmental demands for remodeling, perfusion or permeability is dependent on dynamic changes at adherens junctions, which appear with linear or jagged morphologies 41 (Fig. 4a). In the WT P6 retina, VE-cadherin in arterial ECs formed linear junctions, whereas jagged junctions appeared only on veins and capillaries exhibiting phosphorylated VE-cadherin (Extended Data Fig. 9). YES1 silencing did not prevent formation of jagged junctions but the response to shear stress (3 dyn cm −2 ) was lower than in the control cells (Fig. 4b,c). In contrast, silencing of SRC in HUVECs exhibited a strong inhibitory effect on jagged junctions both in static and flow conditions (Fig. 4b,c). In P6 retinas, the appearance of jagged junctions in the veins of both Yes1 and Src iECKO mice was suppressed but more extensively so in the Src iECKO condition (Fig. 4d,e). These data show that changes in VE-cadherin morphology from linear to jagged does not directly correlate with its phosphorylation status.

VE-cadherin internalization requires Yes, but not Src
In the siYES1-treated, flow-exposed HUVECs, VE-cadherin appeared in broad clusters along the junctions (Fig. 4b). The nature of these clusters was further explored by live cell imaging of HUVECs expressing green fluorescent protein (GFP)-tagged VE-cadherin. In control cells, VE-cadherin was detected in junction-proximal endocytic vesicles, which moved away from the junction. The internalized VE-cadherin vesicles disappeared within 2-5 min (Fig. 5a, top and Supplementary Video 2). In YES1-silenced HUVECs, VE-cadherin instead formed large clusters, which existed for 15-30 min before gradually disappearing, to be replaced by new clusters (Fig. 5a, middle and Supplementary Video 3). In SRC-silenced HUVECs, VE-cadherin endocytic vesicles were observed with similar dynamics as in control cells (Fig. 5a, bottom and Supplementary Video 4).
These data indicated that VE-cadherin internalization was affected by Yes deficiency. VE-cadherin internalization was followed by an antibody feeding assay. Constitutive internalization of VE-cadherin was efficiently inhibited by YES1 silencing, whereas SRC silencing had no effect (Fig. 5b,c). HUVECs cultured under flow (3 dyn cm −2 ) showed increased VE-cadherin internalization, blocked by siYES1 but not siSRC treatment ( Fig. 5d and Extended Data Fig. 10a). Additionally, VEGFA-induced VE-cadherin internalization was arrested by siYES1 but not siSRC ( Fig. 5e and Extended Data Fig. 10b). VE-cadherin endocytic vesicles enriched in pY685 VE-cadherin were detected in vivo, just below the plasma membrane, in vena cava ECs from control mice. The prevalence of endocytotic vesicles was markedly reduced in the Yes1 iECKO vena cava (Fig. 5f,g). 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 the regulation of VE-cadherin endocytosis as binding of p120 masks an internalization motif in VE-cadherin 14 . The presence of the p120-catenin/VE-cadherin complex was assessed by coimmunoprecipitation, which showed similar levels under basal conditions and a trend toward increased complex formation in VEGFA-treated siControl and siYES1 HUVECs (Extended Data Fig. 10c,d). Thus, complex formation with p120 was unaffected by loss of VE-cadherin tyrosine phosphorylation.

Ultrastructure of Yes-deficient endothelial junctions
The decreased VE-cadherin internalization in Yes-deficient ECs could potentially impact junction organization, which was addressed by transmission electron microscopy (TEM) analysis of the mouse ear dermis subjected or not to intradermal injection of VEGFA. In Yes1 iECKO mice, the electron-dense junctional area expected to mark high actin density, known to be disrupted in conjunction with vascular leakage 1,42,43 , was smaller than that in control mice under basal conditions (Fig. 6a,c). While the electron-dense area decreased upon VEGFA injection in WT mice, it increased in the Yes1 iECKO condition (Fig. 6a,c), suggesting disturbed junctional dynamics under Yes1 deficiency. The electron-dense cortical actin area in Src iECKO mice was similar to control mice with a trend toward decreased area with VEGFA treatment (Fig. 6b,d).
In agreement, live imaging of cells labeled with fluorescent actin revealed that actin stress fibers were misaligned and moved toward the periphery in YES1-but not in Control-or SRC-silenced HUVECs (Supplementary Videos 5-8). Moreover, VEGFA treatment increased the relative abundance of cortical actin in YES1-silenced HUVECs but not in SRC-silenced or control cultures. In the siSRC cultures, there was instead an increased proportion of stress fibers relative to cortical actin (Fig. 6e,f) in accordance with the different ultrastructural TEM morphologies of Yes1 iECKO and Src iECKO vessels.

Increased vascular permeability in Yes1 iECKO mice
The consequence of Yes deficiency for vascular barrier function was addressed using several in vivo methods. The basal level of vascular permeability was tested by tail vein injection of Evans blue followed by a 2-h circulation period. Increased levels of Evans blue in the interstitial tissue of the ear skin were detected in Yes1 iECKO mice (Extended Data Fig. 11a). Blood-brain barrier integrity, assessed by the leakage of Evans blue and cadaverine into the cerebral tissue, was unaffected by Yes1 deficiency (Extended Data Fig. 11b,c). Thus, the vascular barrier was compromised in the Yes-deficient dermis and potentially other tissues but not in the CNS.
The permeability of developing retinal vessels was studied by cardiac injection of fixable fluorescent tracer (dextran-TRITC, 10 kDa) in P6 mice. Leakage of the tracer into the perivascular tissue was observed preferentially at the vascular sprouting front, possibly marking an immature vascular barrier, and neither Yes1 deletion nor overexpression changed the number of leakage sites at the sprouting front ( Fig. 7a,b); however, an increased frequency of venous leakage was observed in Yes1 iECKO retinas (eight leakage sites observed in 11 veins from ten retinas) in comparison to the control retinas (two leakage sites in 14 veins from nine retinas) (Fig. 7a,c). In contrast, leakage from veins was not elevated in Yes1 iECOE retinas (Fig. 7c).
Permeability of the mature vasculature was investigated by intravital confocal imaging of adult dermal vessels. In the WT dermis, pVE-cadherin was detected in some but not all venous vessels and, unexpectedly, in some arterioles (Extended Data Fig. 11d). 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 reflect flow patterns that do not conform to the expected high-flow-low-phosphorylation pattern or alternatively, higher expression levels of VE-cadherin. Nevertheless, VE-cadherin phosphorylation in dermal vessels was markedly reduced in adult Yes1 iECKO mice (Extended Data Fig. 11d,e), similar to the findings in the developing retina. To monitor leakage, a fluorescent tracer was injected into the tail vein and leakage from dermal vessels in the ear was monitored after local injection of VEGFA ( Fig. 7d and Supplementary Videos 9 and 10). Leakage was initiated at individual sites in venules and capillaries in both control and Yes1 iECKO mice ( Fig. 7e and Supplementary Videos 11 and 12). Quantification of leakage sites/vessel length showed that the leakage site density was increased in both venules and capillaries in Yes1 iECKO mice (Fig. 7f). The fluorescence intensity profile at leakage sites demonstrated that leakage appeared sooner after VEGFA injection in Yes1 iECKO mice and that leakage was more vigorous in the absence of Yes1 compared to the WT     Article https://doi.org/10.1038/s44161-022-00172-z (Fig. 7g). The number of leakage sites was unaffected in Yes1 iECOE mice compared to WT mice (Fig. 7h), indicating that leakage is not directly regulated by VE-cadherin tyrosine phosphorylation. In contrast to the increased leakage observed in Yes1 deficiency, Src iECKO mice exhibited a reduced number of leakage sites after VEGFA injection (Fig. 7i). The kinetics of the remaining leakage sites was unaffected (Fig. 7j). These results show that while Yes is required for restricting formation of leakage sites at junctions, Src is required for such sites to form. The Y731 phosphosite in VE-cadherin is required for leukocyte extravasation. Leukocyte populations were analyzed by FACS in fluids collected by peritoneal lavage in mice 24 h after saline or VEGFA was   injected. In agreement with the phenotype of Y731F mice 11 , monocyte (CD11b + LY6G − ) extravasation was reduced in VEGFA-treated Yes1 iECKO mice compared to the control (Fig. 7k). Combined, these results 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 inflammatory cells.

Discussion
Regulation of endothelial junctions has been ascribed to Src 44,45 . The viral form, v-Src, localizes to adherens junctions 46 and gap-junction-dependent cell-cell communication is regulated by c-Src 47 . Although broad screens for Src kinase substrates have revealed hundreds of candidate molecules 48 , only a handful have been validated, 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. In agreement, EC-specific loss of Src expression results in impaired cell-matrix adhesion properties 28 .
Even less is known about Yes substrates. Yes phosphorylates the tight junction component occludin in epithelial cells and ECs 49 , but the functional consequence is unclear. Yes has been implicated in negative regulation of the cell cycle by phosphorylating cyclin-dependent kinase 4 (ref. 50 ). Yes moreover plays a role in T-lymphocyte immunity by phosphorylating CD46 (ref. 51 ) and collapsin response mediator protein 2 (CRMP2) 52 ; however, using an in vitro kinase-based screen to identify substrates for SFKs, no specific substrate for Yes was identified 53 . The important role of Yes in EC cell-cell adhesion is suggested by its abundant junctional localization; single-cell transcriptomics shows that Yes1 is more abundant in ECs than in other cell types in a range of mouse tissues (https://tabula-muris.ds.czbiohub.org/) 54 .
A notable conclusion is that the different SFKs must be targeted individually as they serve different, sometimes opposing purposes. Yes deficiency resulted in loss of VE-cadherin phosphorylation, suppressed endocytosis, clustering of VE-cadherin and increased macromolecular leakage. In contrast, in Src deficiency, VE-cadherin phosphorylation was reduced, albeit to a lesser extent, VE-cadherin internalization remained unaffected and macromolecular leakage was suppressed. Combined, these data challenge the cause-consequence relationship between VE-cadherin phosphorylation, internalization and the formation of junctional gaps and macromolecular leakage. Specifically, the loss of the Y685 site has been shown to prevent tracer leakage in healthy and pathological tissues 11,22,55 . Furthermore, VEGFA-or histamine-induced acute vascular leakage in mature dermal vessels is established at pVE-cadherin-high sites in capillaries and postcapillary venules 5,56 . Combined, the data suggest that VE-cadherin phosphorylation is required for its constitutive endocytosis and it sensitizes ECs to rapid VEGFA-induced endocytosis and gap formation at EC junctions (see schematic outline, Fig. 8). Nevertheless, an additional trigger is required for gap formation and macromolecular leakage. Of note, phosphorylation of VE-cadherin could also be a prerequisite for gap closure. Thus, the dysregulated leakage from Yes-deficient 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 localization, allowing phosphorylation of different subcellular pools of VE-cadherin, in addition to the potential involvement of other kinase-specific substrates. Src localizes to focal adhesions in freshly seeded ECs, where it phosphorylates p125FAK 27 . Endothelial FAK-deficiency results in dismantling of adherens junctions 57 , and VEGFA-induced vascular leakage requires FAK-induced phosphorylation of β-catenin 58 , underscoring the important role of matrix adhesion in the regulation of EC junctions. The subcellular localization of cytoplasmic tyrosine kinases is influenced by NH2-terminal fatty acid modification. Both Yes and Src are myristoylated, whereas Yes is also palmitoylated, potentially contributing to steer the subcellular localization and trafficking of Src and Yes 59,60 . Their different subcellular localization would also be compatible with the distinct regulation of the actin cytoskeleton by Yes and Src; Yes-but not Src-deficient HUVECs exhibited an increase in cortical actin and loss of actin stress fibers. Moreover, the electron-dense, cortical actin-enriched junctional area in Yes-deficient dermal vessels was low in the basal condition but increased markedly upon administration of VEGFA. The abnormal actin arrangements in the absence of Yes provide a potential mechanistic underpinning for the changes in junctional dynamics with formation of VE-cadherin clusters and the block in VE-cadherin internalization.
Binding of p120-catenin to a VE-cadherin juxtamembrane motif has been implicated in VE-cadherin stabilization 14 , and in agreement, dephosphorylation of VE-cadherin at Y658 increases the binding affinity for p120 in vitro 19,61 . 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 62 . VE-cadherin/p120 complex formation occurs, however, independently of VE-cadherin phosphorylation 18 . In accordance, p120/VE-cadherin complex formation occurred also in the absence of Yes. Notably, in contrast to certain VE-cadherin endocytosis-deficient mutant models 62 , the expression levels of VE-cadherin and p120 were unaffected by Yes1 deletion. Moreover, YAP/TAZ signaling is implicated in EC adhesion and VE-cadherin internalization 63 ; the potential contribution of Yes to YAP/TAZ signaling warrants further studies.
The exaggerated macromolecular leakage in Yes1 −/− vessels is at odds with the leakage phenotype of Y685F mice, which show suppressed leakage of macromolecules 11,22 and suppressed reverse transmigration of neutrophils 21 . The difference in phenotypes between the Y685F-and Yes-deficient mouse models may be related to the fact that Yes deficiency 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 Y731F vasculature 11 . Thus, although the phosphosites are similarly regulated by Yes, Y658, Y685 and Y731 serve distinct purposes by regulating the extravasation of either macromolecules or inflammatory cells through different mechanisms and potentially at geographically distinct sites in the vasculature.
The new insights presented here deepen the understanding of the complex regulation of adherens junctions, where an interplay between several cytoplasmic tyrosine kinases, such as Yes and Src, orchestrates the process when gaps form and close again and when the vascular surface becomes prepared to dock and expel inflammatory cells.

Mice
The Yes1 flox/flox mouse strain was generated by flanking exon 3 with loxP sites, introduced by homologous recombination on the genetic C57BL/6 background (Taconic/Cyagen; Extended Data Fig. 2a). The generation of Src flox/flox and Cdh5 flox/flox strains has been described 28,55 . These strains were crossed with Cdh5(PAC)-CreER T2 mice (a kind gift from R. Adams) to generate endothelial-specific knockout of Yes1, Src or Cdh5. For analysis in mosaic knockout, the fluorescent reporter mouse line B6.129×1-Gt(ROSA)26Sor tm1(EYFP)Cos /J (stock no.006148, The Jackson Laboratory) was introduced to indicate Cre activity with YFP expression. For retinal EC distribution analysis, the iSuRe-Cre mouse strain (a kind gift from R. Benedito) was crossed with the Yes1 flox/flox mouse. iSuRe-Cre + ; Yes1 wt/wt ; Cdh5CreER T2+ mice were used as control for all analyses of the iSuRe-Cre + ; Yes1 fl/fl ; Cdh5CreER T2+ strain, to ensure that the observed phenotype was not due to Cre toxicity. Conditional Yes1 knock-in mouse H11-CAG-STOP-Yes1 was generated using CRISPR/ CAS (Taconic/Cyagen; Extended Data Fig. 4a). The VE-Cad Y685F and Article https://doi.org/10.1038/s44161-022-00172-z the VE-Cad Y731F mice have been described 11 . Both males and females were included in the experiments. Animal husbandry and procedures were in accordance with institutional guidelines and approved by the Institutional Review Board for animal experimentation at Uppsala University (permit no. 5.8.18-06789/2018).

Intravital vascular leakage assay
Intravital imaging of the mouse ear dermis 5 was performed following systemic administration of 2,000 kDa dextran by tail vein injection. Mice were anesthetized by i.p. injection of ketamine-xylazine (120 mg kg −1 ketamine, 10 mg kg −1 xylazine) to a surgical level, and the ear was secured to a solid support. Mice were maintained at a body temperature of 37 °C throughout the experiment for a maximum 90 min. Timelapse imaging was performed using single-photon microscopy (Zeiss LSM 710) and a high NA water-immersion lens (CF175 apochromat ×25 W NA 1.1, Nikon). For intradermal EC stimulation, a volume of 0.1 µl recombinant mouse VEGFA164 (Peprotech), 100 ng µl −1 was injected using a submicrometer capillary needle. Then, 10 kDa TRITC dextran was used as a tracer for VEGFA. Leakage sites were identified in timelapse imaging following VEGFA injection as defined 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 performed using the PolNet platform 31 . Flat-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 microscopy (Leica SP8) using a ×40 objective. Three-dimensional reconstruction of the vascular network was performed using maximum intensity projection of the vessels illustrated by CD31 immunostaining. A heat map of the relative shear stress level was generated based on the flow simulation by assigning an inlet at the artery and outlet at the vein. Regional average shear stress levels were obtained and plotted combined with the pVE-cadherin levels.

Retinal EC distribution analysis
Chimeric recombination was induced in iSuRe-Cre + mice at P3 by i.p. injection of tamoxifen (100 µg per mouse, Sigma). Retinas were taken at P7 and P15, immunostained for CD31 and flat-mounted. Images were taken by z-stack tile scanning using a ×10 objective on a confocal microscope (Leica SP8). Maximum intensity projection images of whole retinas were used for image segmentation, which was performed with ImageJ resources. The maximum projection of the MbTomato channel threshold was established to distinguish MbTomato + cells from the background. Outliers with a radius between 0.2-1.0 µm were removed. The CD31 channel (after maximum projection) was used to define the outlines of veins and arteries; the optic nerve was used as a mask to define a referential system. For computational analysis, a bespoken Python-based workflow was employed, accessible on GitHub (https://github.com/wgiese/retina-vein-artery-cs). For every pixel in the image, three numbers were computed (using the mask as referential): (1) distance to the nearest vein (d v ); (2) distance to the nearest artery (d a ); and (3) radial distance to the optic nerve (r). From these measures, the relative distances were obtained by ϕ v-a = d v/ (d v + d a ).
The EC distribution was computed by performing the operation for all MbTomato-positive pixels, which were used as a proxy for EC distribution. A kernel density estimation was used to approximate the underlying EC distribution in the two-dimensional coordinate system spanned by ϕ v-a and r. High-degree recombination was induced at P1, P3, P6 and P9 by i.p. injection of tamoxifen (100 µg per mouse per day, Sigma) and retinas were taken at P15 for analysis.

Retinal single-cell RNA-sequencing data
Single-cell RNA-sequencing data generated from WT P6 and P10 retinas 64 was obtained from the Sequence Read Archive (https://www. ncbi.nlm.nih.gov/sra) accession no. SRP322112 and preprocessed with CellRanger (v.5.0.1) using the mm10 reference genome. ECs included in the analysis were defined as positive for Pecam1, Cdh5 or Kdr expression or as having been characterized as ECs in the original publication. Raw counts were normalized with a pooling size factor-based strategy as implemented in scran (v.1.18.7). Subtype annotations were generated using scmap (v.1.16.0) using the cluster assignment described in the original publication. Differential gene expression analysis was carried out using MAST as implemented in Seurat (v.4.1.1). The effects of dropouts in the data were reduced with imputation with the magic Python package (v.0.1.1; k = 9, ka = 3, t = 1) and graphs were generated with ggplot2 (v.3.3.6).

Vessel leakage in postnatal retina
Recombination was induced at P1-3 by i.p. injection of tamoxifen (100 µg d −1 per mouse, Sigma). At P6, mice were anesthetized, and  and Y731. Phosphorylation of VE-cadherin is required for its constitutive internalization, which confers EC junctional plasticity. Junctional plasticity is essential for collective EC migration during angiogenesis as well as for maintaining vascular barrier integrity.
Article https://doi.org/10.1038/s44161-022-00172-z 10 kDa TRITC dextran-lysine (100 µg g −1 body weight, TdB Labs) was injected into the left heart ventricle. After 10 min, eyes were removed and fixed in 4% paraformaldehyde (PFA; Sigma) for 2 h at room temperature (RT). A needle puncture on the cornea facilitated fast fixation. Retinas were immunostained 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.

Transmission electron microscopy
Eight to 12-week-old control or Yes1 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 anesthetized and perfused first with 10 ml Hank's balanced salt solution and then 12 ml cold fixative (1% glutaraldehyde (Sigma), 4% PFA in 0.1 M phosphate buffer) through the left ventricle. Ears were cut and placed in fixative for 30 min at 4 °C, washed three times with PBS, dehydrated and embedded in Epon, LX112 (Ladd Research Industries). Imaging was performed with a Tecnai G2 Spirit 120 kV transmission electron microscope (Fei Europe) and Quamesa CCD camera. Image analysis was done with Fiji processing package of Image J2 software.

Aortic ring assay
Thoracic aortas were taken from P7 mice and cut into 1 mm rings after removal of the connective tissues. Rings were embedded between two layers of rat tail collagen, type I (Thermo Fisher Scientific), and cultivated in Dulbecco's modified Eagle's medium (Gibco) supplemented with 10% fetal bovine serum (FBS) and VEGFA (PeproTech, 30 ng ml −1 ). Then, 4-OH-tamoxifen (5 µg ml −1 , Sigma) was added to the culture for 48 h to induce recombination. After 4 d of initial cultivation, samples were fixed in 4% PFA (Sigma) and analyzed by microscopy after immunostaining.
Proliferation EC proliferation was assessed by i.p. injection of EdU (100 µg per mouse, Thermo Fisher Scientific) at P6. Mice were killed 2 h after injection, and retinas were removed and fixed in 4% PFA (Sigma). EdU staining of retinas was performed using the Click-iT EdU imaging kit (Thermo Fisher Scientific). The samples were co-stained with antibodies against ERG to identify EC nuclei. Images were taken by confocal microscopy (Leica SP8) and analyzed by ImageJ.

Isolation of mouse lung ECs
Mice were treated with tamoxifen (100 µg d −1 per mouse, Sigma) at P1-3. Lungs were taken at P6 and dissociated to obtain cell suspensions using the MACS dissociation kit (Miltenyi Biotec). Isolation of ECs from mouse lungs was performed using anti-CD31 antibody (BD Pharmingen)-conjugated Dynabeads (Thermo Fisher Scientific). RNA from the isolated ECs was extracted immediately by using an RNeasy mini plus kit (QIAGEN).

Cell culture
HUVECs (PromoCell) were cultured at 37 °C, 5% CO 2 in MV2 medium with supplied supplements (PromoCell). Cells were tested for cell morphology and cell-type-specific markers using flow cytometric analyses by the vendor. For VEGFA stimulation, confluent cells were starved in MV2, 0.2% FBS without supplements for 5 h. The same medium with VEGFA165 (50 ng ml −1 , PeproTech) was added and cells were incubated at 37 °C for the indicated time periods.

Yes and Src overexpression in vitro
SRC cDNA was cloned into pLV-CMV-IRIS-PURO-mScarlet plasmid at restriction enzyme site XhoI using In-Fusion Cloning (Takara). YES1 cDNA was cloned into the pLV-CMV-IRIS-PUR-mScarlet plasmid at XhoI site using In-Fusion Cloning (Takara). Lentivirus constructs of control-m-Scarlet, c-Src-mScarlet, and Yes-mScarlet were packaged into lentivirus in HEK-293T cells by co-transfection of third-generation lentiviral packaging plasmids with PEI 2500 (BioScientific). Lentivirus containing supernatant was collected on day 2 and 3 after transfection. Lentivirus was concentrated by Lenti-X concentrator (Clontech, 631232). Transduced target HUVECs were selected with puromycin (1 mg ml −1 ) after 24 h and used for assays after 72 h.

Scratch wound healing assay
HUVECs were seeded at a density of 50,000 cells per well into a 96-well ImageLockTM tissue culture plate (Essen BioScience) and incubated in MV2 medium overnight followed by standardized scratching using WoundMaker (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. The migration speed of one cell was obtained by dividing the migration distance by time.

Confocal live cell imaging
HUVECs were seeded at full confluency 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 2 h. 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 humidified CO 2 incubator (Leica Microsystem). Timelapse z-stack images were scanned every 1 min for 3 h.

In vitro flow treatment
HUVECs were seeded at full confluency on µ-slides (Ibidi) coated with fibronectin (1 µg ml −1 , Sigma) and cultured overnight in degassed MV2 medium. The slides were then connected to the Ibidi pump system to allow degassed medium to flow 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 slides in an incubator for the same amount of time. Cells were fixed in 1% PFA for 10 min for immunofluorescence or lysed in RIPA buffer (Thermo Fisher Scientific) for western blots.

VE-cadherin internalization assay
Control or YES1-silenced confluent HUVECs were incubated with an antibody against the VE-cadherin extracellular domain (Clone BV6, MABT134, Merck Millipore, 10 µg ml −1 ) at 4 °C for 1 h 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 h in the presence of 150 µM chloroquine (Sigma) in MV2 medium with supplements. For VE-cadherin internalization under flow, cells were first cultured under 3 dyn cm −2 flow for 20 h before adding the antibody (Clone BV6, MABT134, Merck Millipore, 10 µg ml −1 ) and chloroquine (150 µM) in the medium followed by continued culturing under flow for 4 h. Cells were then washed for 8 min with PBS containing 25 mM glycine and 3% BSA with Ca 2+ and Mg 2+ at pH 2.7, to remove antibody from the cell surface, followed by rinsing with PBS and fixation with 2% PFA for 10 min at RT. Total VE-cadherin was stained with an antibody generated in another species (AF1002, R&D Systems). Samples were Article https://doi.org/10.1038/s44161-022-00172-z mounted on slides with Fluoromount G (Southern Biotech) and analyzed by confocal imaging.

RNA interference
siRNAs targeting human YES1, SRC or PTPRB (Sigma) were transfected into HUVECs by using Lipofectamine RNAiMAX (Thermo Fisher Scientific). Control cells were transfected with control siRNA that did not target any specific gene. siRNAs were mixed with lipofectamine and 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. The knockdown efficiency was validated by quantitative PCR. YES1 siRNA: SASI_Hs01_00086922; SRC siRNA: SASI_Hs01_00112907; PTPRB siRNA: EHU158501, Sigma; Control siRNA: MISSION siRNA universal negative control (SCI001).

Immunofluorescence
Mice were killed and tissues were removed and fixed in 2% PFA (eyes and ears were fixed for 1 h, vena cava was fixed for 15 min). Samples were incubated in PBS, 0.5% Triton X-100, 1% BSA for 3 h at RT and then with primary antibodies overnight at 4 °C. After three washes at RT, samples were incubated with secondary antibodies overnight at 4 °C, washed three times and mounted on slides for imaging. For staining of cultures, cells were fixed in 1% PFA for 10 min after treatment. After fixation, the cells were washed three times in PBS and permeabilized in PBS, 0.1% Triton X-100 for 5 min, followed by blocking in 1% BSA for 30 min and incubation with primary antibodies overnight at 4 °C. Cells were washed three times in 0.1% Triton X-100, incubated with secondary antibodies for 1 h at RT, washed three times and mounted. Imaging was done using Leica SP8 (Leica Microsystems) confocal microscopes with LAS X software (v.3.5.7.23225). Image processing and analysis were performed using ImageJ software.

Quantitative PCR
RNA was extracted and purified using RNeasy Plus kit (QIAGEN). RNA concentrations were measured by a Nanodrop spectrophotometer (Thermo Fisher Scientific) and adjusted to equal concentrations, followed by reverse transcription using SuperScript III (Thermo Fisher Scientific). Real-time quantitative PCR was performed on a Bio-Rad real-time PCR machine using SsoAdvanced qPCR reagent (Bio-Rad). The housekeeping gene ribosomal protein L19 (Rpl19) was used as an internal control. The comparative Ct method was used to calculate fold differences.

Rho-GTPase activity assay
Control or YES1-silenced HUVECs were starved at 37 °C in basal MV2 medium with 0.2% FBS for 3 h and stimulated with VEGFA164 (50 ng ml −1 , PeproTech) for 15 min. The activities of the Rho GTPases Rac1, CDC42 and RhoA were detected using kits from Cytoskeleton. Briefly, 300 µg of freshly prepared cell lysates were incubated with Rhotekin-RBD (RhoA) or PAK-PBD (CDC42, Rac1) beads for 1 h, 4 °C. Beads were washed five times and boiled in Laemmli buffer with β-mercaptoethanol. Samples were separated by SDS-PAGE (4-12% gradient gel) (Thermo Fisher Scientific), and blotting was performed to detect Rac1, CDC42 and RhoA with antibodies supplied in the kit.

Statistics and reproducibility
Statistical analysis was performed using GraphPad Prism software (9.4.0). Statistical significance in comparisons between two groups was determined by two-tailed Student's t-test or Welch'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 significant at P < 0.05. For animal experiments, no statistical methods were used to predetermine sample sizes. The experiments were not randomized. The investigators were not blinded to allocation during the experiment and outcome assessment.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Article https://doi.org/10.1038/s44161-022-00172-z

Data availability
Full source data for retinal EC distribution analysis is available on Zenodo (https://doi.org/10.5281/zenodo.7229061). Single-cell RNA-sequencing data generated from WT P6 and P10 retinas 64 was obtained from the Sequence Read Archive (https://www.ncbi.nlm. nih.gov/sra), accession no. SRP322112. Additional data supporting the findings in this study are included in the main article and associated files. Source data are provided with this paper.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.