EphrinB2 controls vessel pruning through STAT1-JNK3 signaling

Angiogenesis produces primitive vascular networks that need pruning to yield hierarchically organized and functional vessels. Despite the critical importance of vessel pruning to vessel patterning and function, the mechanisms regulating this process are not clear. Here we show that EphrinB2, a well-known player in angiogenesis, is an essential regulator of endothelial cell death and vessel pruning. This regulation depends upon phosphotyrosine-EphrinB2 signaling repressing JNK3 activity via STAT1. JNK3 activation causes endothelial cell death. In the absence of JNK3, hyaloid vessel physiological pruning is impaired, associated with abnormal persistence of hyaloid vessels, defective retinal vasculature and microphthalmia. This syndrome closely resembles human persistent hyperplastic primary vitreus (PHPV), attributed to failed involution of hyaloid vessels. Our results provide evidence that EphrinB2/STAT1/JNK3 signaling is essential for vessel pruning, and that defects in this pathway may contribute to PHPV.


Introduction
During development and in adult mammals the vessel network expands through angiogenic sprouting into areas with increased need for nutrients and oxygen, and subsequently undergoes complex remodeling through branch pruning, pericyte coverage and basement membrane deposition to generate a quiescent and mature vasculature 1 . Although considerable progress has been made in clarifying the signals that orchestrate endothelial cell sprouting, less is known about the mechanisms controlling blood vessel pruning despite the critical importance of this process to the patterning, density and function of blood vessels. Capillary involution is evident in the hyaloid vessels, which fully regress after providing a temporary blood supply during eye development 2 ; in the primitive retinal vessels, which mature into a stable plexus 3 or degenerate after exposure to hyperoxia 4 ; and in the tumor vasculature, where degenerating vessels border dense and chaotic vasculature 1 . Reduced blood flow 5,6 , VEGF reduction 7,8 , Dll4/Notch activation 3,9 , FYVE expression 10 , exposure to TNFα or IFNγ 11,12 , loss of Nrarp 13 and light-induced responses 14 can provide death signals to the vascular endothelium.
EphrinB2, a transmembrane ligand for Eph receptors that is expressed on arterial endothelium, plays pivotal roles in angiogenesis during development and disease [15][16][17][18] . Genetic experiments in mice have shown that the global inactivation of Efnb2 19,20 , the targeted inactivation of Ephb2 to the endothelium 21 , or replacement of the endogenous gene by cDNA encoding a mutant EphrinB2 that lacks 66 amino acid residues of the cytoplasmic tail 22 similarly impair early embryonic angiogenesis and cause lethality. Since this EphrinB2 cytoplasmic deletion did not impair EphB4 receptor activation, it follows that EphrinB2 intrinsic signaling from the cytoplasmic domain is critical to vascular development 22,23 . Mechanistic studies have revealed that EphrinB2 signaling involving PDZ interactions promotes VEGFR2 activation and angiogenic sprouting, whereas phosphotyrosine-dependent EphrinB2 signaling does not 24,25 . However, EphrinB2 is tyrosine phosphorylated in angiogenic vessels 26 . Genetic evidence has demonstrated that phosphotyrosine-dependent EphrinB2 signaling regulates cell-cell adhesion and cell movement by recruiting Grb4 17 but has not been linked to post-angiogenic vessel remodeling or pruning.
Here we identify a novel pathway controlled by EphrinB2 that is critical for regulation of vessel survival and pruning in the vasculature of the eye. This pathway links phosphotyrosine-dependent EphrinB2 signaling with repression of JNK3 pro-apoptotic activity via STAT1. In the absence of tyrosine-phosphorylated EphrinB2 or JNK3, physiologic involution of hyaloid vessels is impaired producing a syndrome that resembles human persistent hyperplastic primary vitreus (PHPV).

EphrinB2 controls vessel pruning in the eye
To evaluate the contribution of EphrinB2 phosphotyrosine-dependent signaling to vessel pruning of the ocular vasculature, we analyzed knock-in mice with a targeted mutation of the five conserved tyrosine residues (EphrinB2 5Y/5Y mice) in the cytoplasmic tail, which impairs this signaling 23 . The ocular vasculature comprises the hyaloid and retinal vascular systems 27 .
Hyaloid vessels, an arterial vascular network fully developed at birth that supports development of the eye, regress as the retinal vasculature develops 2 . WT hyaloid vessels broadly express tyrosine-phosphorylated EphrinB (p-EphrinB) at postnatal day (p)4, which is expectedly absent from the EphrinB2 5Y/5Y vessels ( Supplementary Fig. 1a). We found that hyaloid vessels in EphrinB2 5Y/5Y mice display significantly reduced branching compared to EphrinB2 WT/WT mice at p3 and p4, vessel thinning and appearance of gaps compromising vessels integrity (Fig. 1a,b). In 3/21 EphrinB2 5Y/5Y mice the hyaloid vessels were segmentally missing and the eyes grossly abnormal ( Supplementary Fig. 1b,c). Type IV collagen immunostaining showed increased regression of hyaloid vessels (collagen IV + CD31 − sleeves) in the EphrinB2 5Y/5Y mice compared to EphrinB2 WT/WT (Fig. 1c-e), whereas endothelial cell proliferation in hyaloid vessels (marked by Ki67) was similarly low (Fig. 1f,g). The number of red blood cells in the hyaloid vessels, was significantly reduced in EphrinB2 5Y/5Y hyaloid vessels compared to control at p3 and p4 (Fig. 2a-c). This red cell reduction was attributable to decreased hyaloid vessel perfusion in EphrinB2 5Y/5Y mice compared to EphrinB2 WT/WT (Fig. 2d-f).
Given that hyaloid vessels supply the developing retina, defective hyaloid vessels could impair retinal and eye development. We found that p4 and p5 EphrinB2 5Y/5Y retinas display abnormally reduced vessel branching in the capillary plexus next to the optic nerve ( Fig.  4a,b), associated with evidence of increased retinal vessel involution at this location (Fig.  4c,d). Endothelial cell proliferation was normal in retinal vessels of EphrinB2 5Y/5Y mice both at the sprouting front and in the capillary plexus proximal to the optic nerve (Fig. 4e,f). We found that p-EphrinB is broadly detected in retinal vessels of EphrinB2 WT/WT mice at p5, and the absence of p-EphrinB selectively marks vessel segments displaying evidence of degeneration ( Supplementary Fig. 3a). Additionally, we found that adult EphrinB2 5Y/5Y retinas display segmental defects in layer structure involving the center and middle regions ( Supplementary Fig. 3b). Together, these results suggested a role of EphrinB2 tyrosine phosphorylation in post-angiogenic regulation of ocular vessels remodeling (Fig. 4g).
These results together with the observation that EphrinB is phosphorylated in cultured HUVEC (Fig. 5d, Supplementary Fig. 3c) indicate that p-EphrinB2 sustains endothelial cell survival. Exogenous FGF2 (Fibroblast Growth Factor-2) and VEGF (Vascular Endothelial Growth Factor-A) did not prevent HUVEC death after EphrinB2 silencing or expression of eB2-5Y (Supplementary Table 1). Since PDZ-dependent EphrinB2 signaling regulates VEGF Receptor 2 (VEGFR2) activity 24 , we tested if defective VEGFR2 function accounted for the failure of VEGF to rescue endothelial cells from death. VEGF similarly stimulated VEGFR2 signaling in control and eB2-5Y-expressing endothelial cells (Fig. 5e). Thus, cell death induced by the phosphotyrosine-deficient EphrinB2 is not attributable to defective VEGFR2 function.
The profound pro-apoptotic effect of eB2-5Y expression in endothelial cells contrasts with the relatively mild vascular phenotype in the EphrinB2 5Y/5Y mice. Since mitigating functions outside the endothelium could contribute to this difference, we focused on pericytes, which promote destabilization/regression of remodeling vessels 26,29 . We find that eB2-5Ymesenchymal cells/pericytes are defective at promoting regression of vascular structures, supporting the integrity of HUVEC-derived vascular structures longer than eB2-WT pericytes (Figure 5f). Given that pericytes appear normally represented along hyaloid vessels of EphrinB2 5Y/5Y mice ( Fig. 3e-g), the mutant pericytes may serve a similar prosurvival function in the globally deficient EphrinB2 5Y/5Y mice mitigating the profound proapoptotic effect of eB2-5Y expression in endothelial cells.
Based on these results, we examined whether the degree of association between p-EphrinB and SHP2 changes during physiological involution of hyaloid vessels. We found that concurrent with a decline in the proportion of hyaloid p-EphrinB + cells (39.9% at p5; 20.8% at p7), p-EphrinB/SHP2 interaction also declines in WT hyaloid vessels from p5 to p7 (Fig.  6k). Most hyaloid vessel involution proceeds in regions intermediate between the center and the periphery at p4-p7 ( Supplementary Fig. 5b,c). Consistent with this, most p-EphrinB/ SHP2 interaction localizes in the peripheral and central regions of hyaloid vessels at p5/p7, where most p-EphrinB + cells reside (Fig. 6j,k, Supplementary Fig. 5b,c). Together, these results suggest that p-EphrinB2 sustains endothelial cell viability by recruiting SHP2, which inactivates STAT1. Consistent with this, SHP2 silencing increased p-STAT1 levels in HUVEC ( Supplementary Fig. 6a). In addition, STAT1 silencing ( Supplementary Fig. 6b) increased HUVEC viability compromized by expression of EphrinB2-5Y (Fig. 7a).
Immunoprecipitation/immunoblotting further revealed that endogenous JNK3 is timedependently phosphorylated in serum-starved HUVEC under conditions that lead to EphrinB2 de-phosphorylation, whereas JNK1 or JNK2 are not phosphorylated (Fig. 7e). In vitro kinase assays further showed that endogenous p-JNK3 induced in HUVEC by serum starvation is biologically active as assessed by phosphorylation of the exogenous target, cJUN (Fig. 7e). Functionally, forced expression of JNK3 promoted HUVEC death (Fig. 7f), and the silencing of JNK3 increased HUVEC viability after EphrinB2 silencing or expression of EphrinB2-5Y ( Fig. 7g; Supplementary Fig. 6c). This indicated that the prosurvival function of EphrinB2in this system is linked to JNK3 repression.
Next we examined whether JNK3 plays a role in the physiological involution of hyaloid vessels. We detected JNK3 in p5 and p7 hyaloid vessels, particularly at sites of vessel degeneration where p-EphrinB is not detected (Fig. 7h, Supplementary Fig. 6d). By measuring JNK3 and p-EphrinB staining intensities in individual cells within hyaloid vessels, we found a time-dependent (from p5 to p7) increase of IB4 + /JNK3 + cells (13.9% to 56.6%) and a decrease of IB4 + /p-EphrinB + cells (39.9% to 20.8%) (Fig. 7i,j). JNK3 high cells generally showed no or low levels of p-EphrinB2 signals (Fig. 7j), and clustered at a mid relative distance between the center and the periphery of hyaloid vessels, where most vessel degeneration occurs at p4-p7 ( Fig. 7k; Supplementary Fig. 5b,c). The p-EphrinB high / JNK3 low cells clustered in the more viable peripheral and central regions of hyaloid vessels (Fig. 7k). Thus, JNK3 marks endothelial cell death in physiologically involuting hyaloid vessels as they lose EphrinB phosphorylation.
We further examined whether p-STAT1 binds to the JNK3 promoter as predicted in silico. Chromatin IP in lysates of HUVEC expressing endogenous p-STAT1 after serum starvation ( Fig. 6e) showed that p-STAT1 binds to the JNK3 promoter region predicted to bind p-STAT1 (Fig. 8a). Maximal p-STAT1/JNK3 promoter interaction temporally coincided with maximal p-STAT1 activation, and was similar in magnitude to that found in IFNγstimulated HUVEC (Fig. 8a). In a reporter assay, WT EphrinB2 significantly reduced JNK3 promoter activity driven by the transfected JAK2 (but not dominant-negative JAK2) plus STAT1, whereas EphrinB2-5Y exerted an insignificant repressive effect (Fig. 8b).
These results support a model (Figure 8c), in which hyaloid vessel regression proceeds principally in regions intermediate between the central and peripheral regions at p4-p5. At this time, hyaloid vessels in the peripheral and central regions do not regress. Nonregressing hyaloid vessels show high-level p-EphrinB, which provides pro-survival signals: p-EphrinB associated with the phosphatase SHP2, which de-phosphorylates STAT1; proapoptotic JNK3 is not active. Instead, regressing hyaloid vessels show low-level p-EphrinB, resulting in increased cell death: p-EphrinB is poorly associated with SHP2; JNK3 is induced and is active. In sum, EphrinB2/JNK3 signaling emerges a key regulator of endothelial cell survival and post-angiogenic vessel remodeling.

JNK3 regulates involution of hyaloid and retinal vessels
A prediction from this model is that JNK3 deficiency reduces physiologic involution of hyaloid vessels. Analysis of JNK3 −/− mice at p5 found that 77% of mice (10/13 mice) from 8 litters showed an abnormal eye phenotype, which affected both eyes in 30% of mice (3/10) and one eye in 70% of mice (7/10). In the 7 most severe cases, the eye and lens were significantly smaller than controls, and that the retina visualized from the rear presented irregular folds/pockets not found in the controls (Fig. 9a,b; Supplementary Fig. 7a,b). The space separating the retina from the lens, which normally contains the vitreous elements, was reduced in size in these JNK3 −/− mice, as it was occupied by a retrolental mass, mostly staining with IB4 (Fig. 9c). p-EphrinB marked viable-appearing IB4 + cells lining the posterior aspect of the lens (arrowheads) in the control and in the JNK3 −/− mice (Fig. 9d). The absence of p-EphrinB in IB4 + cells (arrows) coincided with morphologic evidence of cell degeneration in the controls, but no such evidence of degeneration was noted in the JNK3 −/− IB4 + cells (Fig. 9d). Histologically, the JNK3 −/− retrolental mass appears as an abnormal vascular-type structure, mostly composed of IB4 + cells mixed with NG2 + pericytes, and infiltrated with F4/80 + macrophages (Fig. 9e). This mass is absent from JNK3 +/+ eyes, showing instead normal-appearing IB4 + hyaloid vessels associated with NG2 + and F4/80 + cells (Fig. 9e). The JNK3 −/− retrolental mass (arrows) did not stain for Cleaved caspase-3, which was detected in a proportion of control hyaloid cells (arrowheads), presumably physiologically degenerating (Fig. 9f). The superficial retinal vasculature, normally present at p5 extending outward from the optic nerve, was less extensive in JNK3 −/− retinas ( Supplementary Fig. 7c,d). Instead, clusters of JNK3 −/− hyaloid vessels appeared to form a substitute retinal vasculature, as there was evidence of angiogenic sprouting and filopodia formation at the edge of the JNK3 −/− hyaloid vascular plexus with penetration of the hyaloid vessels into the retinal layer (Supplementary Fig. 7eg). In the 3 less severe cases, the hyaloid vessels displayed significantly greater branching compared to controls at p5 consistent with reduced involution, and the JNK3 −/− eyes were abnormally small.. Additionally, 36% of adult (11)(12)(13)(14)(15)(16) week-old, n=11) JNK3 −/− mice displayed a pathological retrolental mass, consistent with an abnormal persistence of hyaloid vessels (Fig. 9g,h). Thus, JNK3 deficiency causes abnormal persistence of hyaloid vessels and ocular pathology in mice.

Discussion
Here we show that EphrinB2 plays a previously unrecognized role as a critical regulator of endothelial cell survival and death controlling vessel pruning once vessels have formed through angiogenesis. EphrinB2 is an essential mediator of angiogenic sprouting, which drives tip cell guidance and endothelial cell sprouting by promoting VEGFR2 internalization and signaling 15,19,22,24 . In this context, EphrinB2 relies on PDZ interactions for signaling 15,24 . We now discovered that EphrinB2 relies instead on phosphotyrosinedependent signaling to sustain endothelial cell viability, and find that this function is VEGF/ VEGFR2-independent. EphrinB tyrosine phosphorylation is induced by EphB receptors through cell-to-cell interaction 15 and also by FGF2 and Claudins 17 . Since veins, sprouting capillaries 25 , monocyte/macrophages 37 , pericytes 19 and other cells 2438 express EphBs, there is an opportunity for endothelial EphrinB2 activation in vivo. Consistent with this, EphrinB is phosphorylated in the remodeling vasculature of the eye, wounds and tumors 26 . The current results suggest a broader role for EphrinB2 in the vasculature, promoting both vascular sprouting and cell survival through the engagement of distinct signaling pathways.
There is growing evidence describing the importance of a dynamic balance between endothelial cell survival, proliferation and cell death during development of the vascular system and in adult angiogenesis 1,39,40 . The current results show that cessation of EphrinB2 phosphorylation promotes endothelial cell death, suggesting that signals inducing EphrinB de-phosphorylation would ensure endothelial cell death when vessel degeneration is needed. Given that cell death is required for vessel pruning, the identification of signals and molecular pathways modulating EphrinB2 phosphorylation in remodeling vessels is an important advance. Endothelial-intrinsic signals may involve EphrinB2 recruitment of the phosphatase PTB-BL, which de-phosphorylates EphrinB2 41 . Extrinsic signals may originate from endothelium-associated monocytes 37 and pericytes 19 , which express EphB4. Pericytes induce vessels pruning contributing to vessel patterning in the eye 29 , a functional role supported by the current results.
JNK3 is expressed in the brain where it mediates neuronal cell death in various models of stress-or trauma-induced neurodegeneration 34,46,47 , and to a lesser degree in all human tissues 34,48 . We now discovered that JNK3 induces endothelial cell death, and that JAK2/ STAT1 signaling activates JNK3 in endothelial cells. Our results show that hyaloid vessel involution is generally more compromised in mice with JNK3deficiency than with EphrinB2 phosphotyrosine deficiency, perhaps attributable to compensatory pathways downstream of EphrinB2 signaling. Remarkably, our study links JNK3-deficiency in mice with PHPV in humans attributed to the abnormal persistence of hyaloid vessels 35,49 . Previously, deficiencies of Angiopoietin-2 50 , p53 51 , Arf (alternative reading frame) 52 , frizzled-5 53 and Bax and Bak 54 were linked to PHPV development. Whether these deficiencies may be coupled to JNK3 is not clear, but JNK signaling regulates p53 stability 55 and ARF nuclear translocation 56 .
To our knowledge, STAT1 signaling has not been linked to JNK3 regulation, despite the importance of IFNγ-induced STAT1 activation in promoting cell death 57 . Here we show that p-STAT1 induced by IFNγ targets the JNK3 promoter in endothelial cells. Besides potentially contributing to understanding of inflammation-induced vascular damage 58 , this new EphrinB/STAT1/JNK3 pathway suggests that post-angiogenic endothelium in the eye and in other tissues such as tumors, which express tyrosine-phosphorylated EphrinB, survives at least in part via repression of JNK3 pro-apoptotic signals. The fully developed, resting vasculature does not usually express active EphrinB, suggesting that other mechanisms must sustain vessel integrity. Many biochemical aspects underlying maturation of nascent vessels are not yet clear 1 . The current results suggest that transition from EphrinB2 phosphotyrosine dependency to independency is a critical step in physiological vessel maturation.
Since dephosphorylation of EphrinB2 is expected to lead to accelerated endothelial cell death and vessel regression, the current results disclose novel experimental approaches for inducing vessel pruning. Src family kinases phosphorylate EphrinB2 41,59 , JAK inhibitors reduce STAT activation and SHP phosphatases inactivate STAT1. Thus, tyrosine kinase inhibitors targeting Src family kinases 60 , SHP inhibitors 61 and JAK inhibitors 62 hold promise for pharmacologically pruning vascular beds that either fail to undergo physiological involution such as hyaloid vessels in PHPV, or contribute to disease progression as it is observed in certain cancers types.

Reporter assays
A transfection-ready Gluc-ON ™ dual-reporter vector system pEZX-PG04 that uses Gaussia Luciferase (GLuc) as the promoter reporter upstream JNK3 promoter sequences and SEAP (secreted Alkaline Phosphatase) as the internal control for normalization (GeneCopoeia no. HPRM23021-PG04) was included in co-transfection mixtures with expression plasmids for JAK2 (a gift of I. Daar, NCI, NIH) and STAT1 (Addgene, ID no. 12301) or DN-JAK2 (A gift of I. Daar, NCI, NIH) and STAT1 alone, with EphrinB2-WT or with EphrinB2-5Y. Supernatants (obtained 48-72 hr after HEK293T transfection) were monitored for Gaussia luciferase and SEAP activities, using Secrete-Pair ™ Dual Luminescence Assay Kit (GeneCopeia, no SPDA-D010). Optima luminometer was used to inject 100 μl of Gaussia Luciferase and SEAP Assay Kit substrate, and light emission (in relative light units, RLUs) was measured according to manufacturer instructions.

Immunofluorescence
Endothelial cells grown on gelatin-coated glass chamber slides were fixed in 4% paraformaldehyde (PFA). Mouse eye balls were fixed with 2% or 4% PFA overnight or 4hr at 4°C, and the sclera was removed under stereomicroscopy (Stemi SV11, Carl Zeiss). Hyaloid vessels and retinas were prepared from the retinal cup 63 , which was blocked and permeabilized with blocking buffer (PBS containing 1% Triton X-100, 10% FBS, 10 mM glycine, 0.1 mM CaCl 2 and 0.1 mM MgCl 2 , or PBS containing 0.1% Tween20 and bovine serum) for 1 hour. Paraffin-embedded human persistent hyperplastic primary vitreous (PHPV) specimens (from anonymized excess tissue not required for diagnosis acquired with approval of Johns Hopkins IRB; protocol: "Pathological analysis of ocular trauma and other non-neoplastic eye diseases") were sectioned with a microtome (Leica RM2155, Leica Biosystems) at 10 μm. The sections were deparaffinized with xylene and ethanol, and then re-fixed with 4% PFA on a glass slide TRUBOND 360 (Tru Scientific, no. 5079W) for 20 min at room temperature. The sections were treated with Uni-Trive solution (Inovex Bioscience, no.

Image quantification
The number of vessel branch points (Figures 1b, 4b) and joints (Figure 5f) was counted with ImageJ software (NIH). 26 To measure branch points of hyaloid vessels (Supplementary Figure 5c), we used four images each from 4 whole hyaloid vessels at p4, p5 and p7 stained with IB4. The image of the whole hyaloid vessels was divided in 10 regions based on relative distance from the central optic nerve to the periphery using ImageJ. The area was measured by ImageJ, and converted pixels into mm 2 . The number of branch points in each region was measured by ImageJ and was divided by the area providing the average number of branch points of hyaloid vessels/mm 2 in each region. To measure Collagen IV + /CD31 − sleeves in hyaloid and retinal vessels (Figures 1e and 4b), we used four images each from 4 whole hyaloid vessels at p4 stained with anti-collagen IV and anti-CD31. In Figure 4d, 4 fields of retina were used for this analysis. We used ImageJ with "Segmented line tool and ROI manager"to measure the length of collagen IV + CD31 − sleeves. The average length of collagen IV + CD31 − sleeves in whole hyaloid vessels is shown. In Figure 4d, the length of collagen IV + CD31 − sleeves was divided by the length of CD31 + vessels and the average ratio was shown. To measure Ki67 + CD31 − cells in hyaloid and retinal vessels (Figures 1g  and 4f), we counted by ImageJ using "Analyze Particle tools" the nuclei of Ki67 + CD31 + cells and the total number of vascular CD31 + cells in hyaloid and retinal vessels. The percentage of Ki67 + CD31 + cells in CD31 + vascular cells was calculated and the average ratios were shown. To count the number of red blood cells in hyaloid vessels (Figure 2b,c), hyaloid vessels were stained with biotinylated IB4, and then stained with AlexaFluor 546labeled streptavidin. To enhance autofluorescence of red blood cells in hyaloid vessels, flattened hyaloid vessels were fixed with 1% glutaraldehyde/3.8% PFA in PBS for 20 min at room temperature. Images were obtained with Carl Zeiss LSM 780 (Carl Zeiss) microscope using ZEN software (Carl Zeiss). Excitation was induced with argon laser (488 nm) and HeNe laser (561 nm), and emission was detected for green fluorescence at 495-535 nm and for red fluorescence at 565-600 nm. Green and red autofluorecence-positive cells were counted as red blood cells using Image J software (NIH). The length of hyaloid vessels was measured by Image J software. The number of red blood cells was divided by the length of hyaloid vessels. To measure vessel perfusion with FITC-dextran (Figure 2f), fourty μl PBS containing FITC-dextran (2000000 mol wt, Sigma Aldrich no. FD2000S, 50 mg/ml) and poly-L-Lysine (300 kDa, Sigma no. P-1524, 10 mg/ml) were injected retro-orbitally into p4 littermate EphrinB2 WT/WT and EphrinB2 5Y/5Y mice. After 5 min, the pups were euthanized, eyes collected and fixed with 4% PFA overnight at 4°C. Sclera was removed under a stereomicroscope. Hyaloid vessels were prepared, stained with IB4 and flattened with 4% low-melting temperature agarose in DPBS 63 . Images were acquired with a Zeiss LSM780 microscope (Carl Zeiss). The total intensity of FITC-dextran in whole hyaloid vessels and the total length of whole hyaloid vessels were measured by ImageJ. The total intensity of FITC-dextran was divided by the length of hyaloid vessels. The results are shown as the average intensity of FITC-dextran/mm hyaloid vessel.
To measure cleaved caspase-3 intensity in endothelial cells and macrophages (Figure 3c,d), hyaloid vessels at p4 were stained with anti-cleaved caspase-3, IB4 and DAPI. Images were acquired with a Zeiss LSM780 microscope. Region of interest (ROI) setting for macrophages and endothelial cells was set by ImageJ Using the ROI, total intensity was divided by the area of individual cells. Mean intensity of cleaved caspase-3 in individual cells was plotted in Figures 3c,d. Individual dots are representative of single cells.
For quantitation of JNK3 and p-EphrinB fluorescence intensity in individual endothelial cells (Figure 7i,j) ROIs were generated by ImageJ software (NIH) based on DAPI staining of hyaloid vessels. The ROIs were applied to images of hyaloid vessels stained for JNK3 and p-EphrinB. After applying the ROIs, total intensity of JNK3 and p-EphrinB, and area in each ROI were measured by ImageJ. Mean intensity of JNK3 and p-EphrinB in each ROI was calculated by dividing the total intensity of JNK3 and p-EphrinB by the area of each ROI. In Figure 7i, the mean intensities of JNK3 and p-EphrinB staining in each ROI are shown as a dot plot.
To measure the fluorescence intensity of JNK3 and p-EphrinB in each region of hyaloid vessels (Figure 7k), an image of hyaloid vessels was divided into 10 regions based on relative distance from the center (optic nerve) to the periphery using ImageJ. Fluorescence intensity of JNK3 and p-EphrinB in each region of hyaloid vessels was measured by ImageJ, and was subsequently normalized by DAPI fluorescence intensity in each region to adjust for different cell number in each region (there is a cell number gradient from the center to the periphery of hyaloid vessels).
For 3D reconstitution and surface rendering of sprouting hyaloid vessels in retina (Supplementary Figure 7), Z stack images were acquired by Zeiss LSM780 from 0.5 μm thickness/layers and processed using Imaris software (Bitplane Scientific software).

Proximity ligation assay
PLA was used to visualize proximity colocalization (<40 nm) of EphrinB2+JAK2, EphrinB2+STAT1, and EphrinB2+SHP2 in HUVEC and hyaloid vessels using Duolink Detection kit (Olink Bioscience, Uppsala, Sweden). The cells were fixed with 4% PFA for 20 min at room temperature. Flattened mouse hyaloid vessels were embedded in 4% hydrogel solution (4% acrylamide, 0.05% bis-acrylamide, 0. After washing, cells, hyaloid vessels and sections were incubated with secondary antibodies with PLA probes (MINUS probe-conjugated antirabbit IgG+PLUS probe-conjugated anti-mouse IgG for JNK3, EphrinB2+STAT1 and EphrinB2+-STAT1 detection; PLUS probe-conjugated anti-goat IgG+MINUS probeconjugated anti-rabbit IgG for detection of EphrinB2+JAK2, EphrinB2+STAT1, and EphrinB2+SHP2; Olink Bioscience, Uppsala, Sweden). Circularization and ligation of the oligonucleotides in the probes was followed by an amplification step. A complementary fluorescent-labeled probe was used to detect the product of rolling circle amplification. The cells and the sections of human PHPV were stained with AlexaFluor 488-Zenon-labeled mouse anti-human CD31 (Covance/Biolegend, no. SIG-3632). Slides were mounted with Duolink II Mounting Medium containing DAPI. Images were obtained with Carl Zeiss LSM 710NLO (Carl Zeiss) or LSM 780 (Carl Zeiss) using ZEN software (Carl Zeiss). In HUVEC, the number of PLA dots was counted using Image J software (NIH). Quantifications are given as mean±SD. Hyaloid vessels were divided into 10 regions based on distance from center (optic nerve) to periphery. Total intensity of PLA, total intensity of DAPI staining and total area was measured by Image J software in the each region. Total intensity of PLA signal in each region was divided by area (= total intensity/mm 2 ). Hyaloid vessel density differs in different regions. Total intensity of PLA signal/mm 2 was normalized by total intensity of DAPI staining.

Proximity extension assay
We applied PEA 66 to detect protein complexes in lysates of hyaloid vessels. Lysates were prepared from hyaloid vessels of EphrinB2 WT/WT and EphrinB2 5Y/5Y mice at postnatal day 5 with lysis buffer (25 mM Tris-HCl, pH7.4, 137 mM NaCl, 3 mM KCl, 0.5% IGEPAL CA-630, 1 mM sodium orthovanadate, 20 mM sodium fluoride, 25 mM sodium pyrophosphate, 50% glycerol, proteinase inhibitor cocktail). Lysates are stored at -20°C until use. Protein concentration in the lysate was adjusted to 0.5 mg/ml. Antibodies were labeled with Proseek Probemaker (Olink Bioscience); anti-EphrinB2 rabbit monoclonal antibody (Abcam, no. 150411) was labeled with oligo DNA probe A; anti-SHP2 rabbit monoclonal antibody (Cell Signaling Technologies, no. 3397), anti-JAK2 rabbit monoclonal antibody (Cell Signaling Technologies, no. 3230) and anti-STAT1 rabbit polyclonal antibody (Cell Signaling Technologies, no. 9172) were labeled with oligo DNA probe B according to the manufacturer's protocol; PEA was performed using Proseek Assay Development Kit (Olink Bioscience) according to the manufacturer's protocol. Briefly, probe A-and probe B-labeled antibodies were diluted to 65 pM with assay solution, and then mixed with the lysate. After incubation, the region annealed by probe A and probe B (based on proximity colocalization of the proteins in the lysate) was extended by DNA polymerase. The extended DNA product was quantified by real-time PCR with a specific TaqMan probe. The lysis buffer without hyaloid vessels was used as a negative control for PEA. The negative control was used as background. Fold increase of each complex was calculated based on Ct value of the negative control by the comparative Ct method. Input of SHP2, JAK2 and STAT1 content was measured by Western blotting using the same lysate used in the PEA.