Endothelial HIF2α suppresses retinal angiogenesis in neonatal mice by upregulating NOTCH signaling Free

Intracellular oxygen is sensed by a family of evolutionarily conserved prolyl hydroxylase domain (PHD) proteins, also known as EGLNs (Bruick and McKnight, 2001; Epstein et al., 2001; Ivan et al., 2002). In mammals, there are at least three closely related PHD family members, including PHD1 (EGLN2), PHD2 (EGLN1) and PHD3 (EGLN3). Among them, PHD2 is most broadly expressed (Epstein et al., 2001; Takeda and Fong, 2007; Takeda et al., 2006).

PHDs use molecular oxygen as a substrate to hydroxylate specific prolyl residues in hypoxia-inducible factor (HIF) α proteins, including HIF1α and HIF2α, the latter of which is also named endothelial PAS 1 (EPAS1) (Epstein et al., 2001; Huang et al., 2002; Tian et al., 1997; Wang and Semenza, 1995). Once hydroxylated, HIFα proteins undergo pVHL- (von Hippel-Lindau protein) and E3 ligase-dependent polyubiquitylation, followed by rapid proteasomal degradation (Epstein et al., 2001; Ivan et al., 2002; Maxwell et al., 1999). In hypoxic or PHD-deficient tissues, HIFα proteins accumulate to high levels because prolyl hydroxylation is blocked (Jiang et al., 2021; Semenza, 2019; Takeda et al., 2014).

In mice, the development of the primary retinal vascular bed begins at birth (post-natal day 0 or P0) and is complete by P7-P8. Morphologically, nascent vascular branches emerge at the rim of the optic nerve head (ONH) and expand in radial directions until the vascular front (VF) reaches retinal periphery after 7 to 8 days (Duan et al., 2017; Fruttiger, 2002). This process depends on a VEGFA gradient established by hypoxic astrocytes ahead of the VF (Gerhardt et al., 2003; Rattner et al., 2019). Because it is located on the inner retinal surface, the primary vascular bed is also referred to as the superficial plexus (SP).

It is unknown how oxygen-sensing mechanisms in retinal endothelial cells (ECs) regulate angiogenesis, although their roles in vascular growth have been characterized in several other cell and tissue types. For example, HIF2α maintains retinal astrocytes in immature states, which are proangiogenic, whereas PHD2 promotes the differentiation of retinal astrocytes towards non-angiogenic mature states (Duan and Fong, 2019; Duan et al., 2014b; Perelli et al., 2021). In contrast to HIF2α, HIF1α is dispensable in retinal astrocytes (Duan et al., 2014b). In non-retinal tissues, such as certain tumors, EC-specific deletion of Hif1alpha (Hif1a) or Hif2alpha (Epas1) suppresses angiogenesis (Skuli et al., 2009, 2012; Tang et al., 2004). In the mouse hindlimb ischemia model, endothelial HIF2α deficiency suppresses arteriogenesis (Skuli et al., 2009, 2012; Tang et al., 2004).

Mechanistically, HIF2α promotes arteriogenesis by upregulating delta-like 4 (DLL4), a membrane-bound NOTCH ligand (Skuli et al., 2012). However, it should be noted that the role of DLL4 is context dependent. Although it positively regulates hindlimb arteriogenesis and tumor angiogenesis (Skuli et al., 2009, 2012), DLL4 strongly inhibits retinal vascular development (Benedito et al., 2009; Hellström et al., 2007; Jakobsson et al., 2010; Suchting et al., 2007; Williams et al., 2006).

In this study, we have investigated the relationship between endothelial oxygen sensing, NOTCH signaling and retinal angiogenesis. Our data demonstrate that HIF2α accumulation in PHD2-deficient retinal ECs may severely suppress, instead of promote, vascular growth, in part by upregulating DLL4 expression and NOTCH signaling. A corollary of this conclusion is that under physiological conditions, oxygen-dependent endothelial HIF2α degradation is supportive of retinal angiogenesis.

EC-specific deletion of floxed Phd2 (Egln1) was mediated by Cdh5(Pac)Cre^ERT2 (Wang et al., 2010). At P6, Phd2 mRNA levels in retinal ECs of tamoxifen-treated Phd2^f/f/Cdh5(Pac)Cre^ERT2 (Phd2^EC−/−) mice was 18.2% of normal levels (Fig. S1A). To examine retinal vascular morphogenesis, whole-mount retinas from P6, P8 and P12 mice were stained with Alexa Fluor 594-isolectin B₄ (IB₄). Laser confocal imaging revealed that the vascular front (VF) advanced more slowly in Phd2^EC−/− mice, although it did eventually reach the retinal periphery (Fig. 1A-G). Furthermore, Phd2^EC−/− mice displayed reduced junctional points and microvascular density (Fig. S1A-F,H,I).

Although retinal vascular development was reported to be suppressed in tamoxifen-treated Cdh5(Pac)Cre^ERT2 mice (Brash et al., 2020), we did not observe the same phenomenon in Ai9/Cdh5(Pac)Cre^ERT2 mice, where Ai9 is a transgene encoding a Cre-inducible TdTomato reporter (Fig. S2). It should be noted, however, that the mice analyzed by Brash et al. were in C57BL/6 strain background, whereas those used in our study had a mixed background of C57BL/6 and CD1.

We also investigated how retinal vascular development might be affected by EC-specific HIFα deficiency. Tamoxifen-treated mice of three different genotypes, including Hif1alpha^f/f/Cdh5(Pac)Cre^ERT2 (Hif-1α^EC−/−), Hif2alpha^f/f/Cdh5(Pac)Cre^ERT2 (Hif-2α^EC−/−) and Hif1alpha^f/f/Hif2alpha^f/f/Cdh5(Pac)Cre^ERT2 ({Hif-1α/2α}^EC−/−), displayed grossly normal retinal vascular morphogenesis (Fig. S3), despite highly efficient deletion of Hif1alpha and/or Hif2alpha in neonatal retinal ECs (Fig. S1). These findings indicate that endothelial HIF1α and HIF2α are dispensable for retinal vascular development. For reference, formal genotype nomenclature and corresponding nicknames are compared in Table S1.

To identify the cellular basis of defective retinal vascular development in Phd2^EC−/− mice, we evaluated tip cell development and EC proliferation. Tip cells were examined in z-stack confocal images of flat-mount retinas, double-stained with IB₄ and anti-ERG (ETS-related gene). Anti-ERG staining labels nuclei in endothelial cells (Shah et al., 2017; Yanagida et al., 2020), whereas IB₄ binds to proteoglycans carrying alpha-D-galactosyl residue, which are enriched on EC surfaces (Laitinen, 1987). Tip cells were identified as IB₄⁺ cells located at the VF, containing large ERG⁺ nuclei, and displaying prominent filopodia structures (exemplified in the expanded view under Fig. 2A). In the course of this analysis, we noticed that retinas from Phd2^EC−/− mice contained a significant number of IB₄⁺/ERG⁺ ECs that were very similar to tip cells when viewed at low magnifications but displayed no or only blunted filopodia at higher magnifications (compare Fig. 2A with 2B, with enlarged views below main panels). Therefore, we coined the term ‘vascular front cells’ (VFCs), which includes both normal and defective tip cells.

Tip cells and VFCs were quantified per 0.8 mm length along the VF. Relative to floxed controls, the number of normal tip cells was significantly reduced in Phd2^EC−/− mice (Fig. 2A,B,E). However, the number of VFCs was comparable between Phd2^EC−/− mice and floxed controls (Fig. 2A,B,F). Defective tip cell structures in Phd2^EC−/− mice may partially explain retarded vascular expansion towards the retinal periphery.

EC proliferation was examined by monitoring the incorporation of 5′-bromo-2′-deoxyuridine (BrdU) into replicating DNA. At P6, neonatal mice were injected with BrdU, and retinas were isolated for whole-mount double staining with IB₄ and anti-BrdU. Proliferating ECs were quantified as the number BrdU⁺ ECs per mm length of IB₄⁺ blood vessels (BVs). Relative to floxed mice, Phd2^EC−/− mice displayed significantly reduced EC proliferation (Fig. 2C,D,G).

Retinal vascular development is a dynamic process involving not only growth but also regression, which is important for the retinal vascular pattern to transform from a nascent honeycomb pattern into a mature vascular tree. On the other hand, excessive microvascular regression may disrupt normal development. We investigated microvascular regression by two complementary methods: detection of apoptotic ECs and identification of regressing capillaries. To reveal apoptotic retinal ECs, whole-mount retinas were double-stained with IB₄ and using anti-caspase 3 (active fragment), followed by flat-mounting and laser confocal imaging. CAS3⁺/IB₄⁺ cells were present sporadically in Phd2^f/f mice but abundantly in Phd2^EC−/− mice (Fig. 3A-F,I). The CAS3⁺ signals displayed a peculiar continuous pattern in the retinal vasculature of Phd2^EC−/− mice. However, the existence of such a pattern is unlikely due to a technical issue, because the same antibody revealed individual, non-continuous CAS3⁺ retinal ECs in positive controls (retinas from P7 wild-type mice briefly exposed to 75% oxygen, Fig. S4).

Nonetheless, the excessive abundance of CAS3⁺ retinal ECs does raise the question whether all of these cells were undergoing active apoptosis. Therefore, we complemented the study using two additional approaches. First, we quantified regressing capillaries, which were identified as IB₄⁻ vascular remnants that retained Collagen IV⁺ (Col IV⁺) basement membrane. At P6, IB₄⁻/Col IV⁺ vascular segments were found more frequently in Phd2^EC−/− mice than in floxed controls (Fig. 3G,H,J), indicating that PHD2 deficiency in retinal ECs led to increased microvascular regression. Furthermore, we carried out TUNEL staining. As shown in Fig. 4, TUNEL⁺ signals in the superficial plexus were increased in Phd2^EC−/− mice relative to floxed mice.

We have previously shown that PHD2 deficiency leads to the accumulation of HIF1α and HIF2α proteins in a variety of embryonic, neonatal and adult tissues (Duan et al., 2011, 2014a; Takeda et al., 2014, 2006). Therefore, we evaluated HIF1α and HIF2α proteins levels in the retinal vasculature of Phd2^EC−/− mice. By double-staining whole-mount retinas with IB₄ and using either anti-HIF1α or HIF2α, we demonstrated the accumulation of both proteins in retinal blood vessels of Phd2^EC−/− mice (Fig. 5).

To determine whether elevated HIF1α or HIF2α protein levels were responsible for suppressed retinal angiogenesis in Phd2^EC−/− mice, we carried out EC-specific disruption of Hif1alpha and Hif2alpha in Phd2^EC−/− background. Technically, this was accomplished by treating newborn Phd2^f/f/Hif1alpha^f/f/Cdh5(Pac)Cre^ERT2 and Phd2^f/f/Hif2alpha^f/f/Cdh5(Pac)Cre^ERT2 mice with tamoxifen, resulting in mice nicknamed {Phd2/Hif-1α}^EC−/− and {Phd2/Hif-2α}^EC−/−, respectively.

At P8, whole-mount retinas were stained with IB₄, flat-mounted and analyzed by confocal imaging. Retinal vascular development was defective in {Phd2/Hif-1α}^EC−/− mice, displaying slower VF advancement towards the retinal periphery and inefficient branching morphogenesis (Fig. 6A,B,I,J). By contrast, retinal vascular morphogenesis appeared normal in {Phd2/Hif-2α}^EC−/− mice (Fig. 6C,D,I,J). These data demonstrate that elevated HIF2α, but not HIF1α, was responsible for suppressed retinal angiogenesis in Phd2^EC−/− mice.

At P6, {Phd2/Hif-1α}^EC−/− mice had reduced number of BrdU⁺ retinal ECs than their floxed controls (Fig. 6E,F,K). In {Phd2/Hif-2α}^EC−/− mice, however, BrdU⁺ retinal ECs were similarly abundant to their floxed controls (Fig. 6G,H,K). Tip cell formation was also rescued in {Phd2/Hif-2α}^EC−/− mice (Fig. S5).

CAS3⁺/IB₄⁺ ECs were abundant in {Phd2/Hif-1α}^EC−/− mice but rare in {Phd2/Hif-2α}^EC−/− or floxed mice (Fig. 7), indicating that EC-specific deletion of Hif2alpha but not Hif1alpha rescued retinal EC survival in Phd2^EC−/− background. In a complementary analysis, regressing microvascular segments were detected as IB₄⁻/Col IV⁺ structures. IB₄⁻/Col IV⁺ vascular segments were infrequent in {Phd2/Hif-2α}^EC−/− mice (Fig. S5), which was in sharp contrast to their frequent occurrence in Phd2^EC−/− mice (Fig. 3H,J). Overall, these data demonstrate that retinal vascular morphogenesis was rescued in {Phd2/Hif-2α}^EC−/− mice and suggest that defective retinal vascular development in Phd2^EC−/− mice was due to HIF2α accumulation in retinal ECs.

Because the expression of DLL4, a NOTCH ligand, was reported to be upregulated by HIFα proteins in ischemic hindlimb tissues, certain tumors and cultured ECs (Feng et al., 2016; Skuli et al., 2009, 2012), we examined whether it was also upregulated in the retinal vasculature of Phd2^EC−/− mice. Briefly, retinas were isolated at P4 and double-stained with IB₄ and anti-DLL4 (Fig. 8A-F), which revealed colocalization of IB₄⁺ and DLL4⁺ signals. Quantification of the ratio between DLL4⁺ and IB₄⁺ pixel values demonstrated increased DLL4 expression in PHD2-deficient retinal blood vessels (Fig. 8G). The same conclusion was reached by qRT-PCR analysis of total RNA from retinal ECs (Fig. 8H).

To determine whether DLL4 upregulation was HIF2α dependent, we compared DLL4 expression between retinal ECs of Phd2^EC−/− and {Phd2/Hif-2α}^EC−/− mice. Briefly, eye cryosections were prepared at P5, double-stained with IB₄ and anti-DLL4, and imaged by confocal microscopy. In the superficial plexus (SP), DLL4 was upregulated in Phd2^EC−/− but not {Phd2/Hif-2α}^EC−/− mice (Fig. 9A-L,Y). Furthermore, the expression of HEY2, a NOTCH downstream target (Fischer and Gessler, 2003), was also selectively upregulated in Phd2^EC−/− but not {Phd2/Hif-2α}^EC−/− mice (Fig. 9M-X,Z).

DLL4 and HEY2 staining signals were also elevated in the retinal region corresponding to INL/IPL area of Phd2EC^−/− mice. Although unexpected, several features of the staining patterns indicate that they reflect the upregulation of DLL4 and HEY2 in these retinal areas, rather than anti-DLL4 and anti-HEY2 antibodies nonspecifically adhering to unrelated proteins. First, staining signals are limited to INL/IPL region and are not found in the rest of the retinal parenchyma. Second, these signals are genotype specific, appearing only in Phd2EC^−/− mice. Third, DLL4- and HEY2-positive signals are located in the same retinal area. Nonspecific background staining by two different antibodies would have resulted in more random staining patterns.

To further analyze Dll4 and Hey2 expression, we performed qRT-PCR analysis of RNA isolated from retinal ECs. Data shown in Fig. 9AA,BB demonstrate that at P6, Dll4 and Hey2 mRNA levels were increased in Phd2^EC−/− but not {Phd2/Hif-2α}^EC−/− mice. Taken together, these data demonstrate that the NOTCH signaling pathway is activated in retinal ECs of Phd2^EC−/− mice through HIF2α-dependent DLL4 upregulation.

We also analyzed the expression of Vegfa, a prototypic HIFα target gene. At P5, VEGFA protein levels were upregulated in the retinal superficial plexus of Phd2^EC−/− mice relative to floxed controls (Fig. 10A-G). In addition, qRT-PCR analysis also demonstrated increased Vegfa mRNA levels in retinal ECs of Phd2^EC−/− mice (Fig. 10H).

In neonatal mice, the DLL4-NOTCH signaling pathway is known to suppress retinal vascular development (Hellström et al., 2007; Wei et al., 2013). Therefore, we asked whether excessive NOTCH signaling might be responsible for diminished retinal angiogenesis in Phd2^EC−/− mice. To answer this question, we treated neonatal Phd2^EC−/− mice with DAPT [(N-{N-(3,5-DifluorophenacetylL-alanyl)}-S-phenylglycine t-butylester], which suppresses NOTCH signaling by inhibiting γ-secretase (Dovey et al., 2001; Geling et al., 2002).

In published studies, intraperitoneal injection of 200 mg/kg of DAPT caused excessive retinal angiogenesis in wild-type neonatal mice (Benedito et al., 2009; Suchting et al., 2007). To minimize effects in wild-type (floxed) background, we treated Phd2^f/f and Phd2^EC−/− mice with DAPT at significantly reduced dosage and frequency. Under such conditions, DAPT did not significantly affect retinal angiogenesis in floxed controls (Fig. 11A,B). In Phd2^EC−/− mice, however, DAPT partially rescued retinal angiogenesis (Fig. 11C,D,G,H). In a second approach, we inhibited NOTCH signaling with DLL4 blocking antibody. Intravitreal injection of anti-DLL4 but not control IgG partially rescued retinal angiogenesis in Phd2^EC−/− mice (Fig. 11E,F,I,J).

To validate that NOTCH signaling was indeed inhibited by DAPT and DLL4-blocking antibody, we quantified the expression of HEY2. Double-staining of eye cryosections demonstrated that DAPT or anti-DLL4 suppressed HEY2 expression in the superficial plexus of Phd2^EC−/− mice (Fig. S6). These findings support the notion that the DAPT and anti-DLL4 promoted retinal vascular development in Phd2^EC−/− mice by inhibiting NOTCH signaling.

The oxygen-induced retinopathy (OIR) model, also known as the mouse model for retinopathy of prematurity (ROP) (Pierce et al., 1995), is widely used to study vascular growth in ischemic retinal tissues. In phase I of the model, neonatal mice are exposed to 75% oxygen for 5 days from P7 to P12, resulting in significant microvascular obliteration in the central half of the retina. In phase II, oxygen-exposed mice are returned to ambient room air for 5 more days. During this period, retinal tissues experience hypoxia due to poor perfusion, causing retinal ischemia and neovascularization. A hallmark of ischemic retinal neovascularization is the formation of neovascular tufts which protrude into the vitreous cavity. Meanwhile, regenerative angiogenesis also occurs, characterized by the growth of peripheral microvessels towards the avascular central area.

We applied this model to Phd2^f/f and Phd2^EC−/− mice, which were treated with tamoxifen at P4-P6 (deletion efficiency is shown in Fig. S1F). At the end of phase I, both Phd2^f/f and Phd2^EC−/− mice had large avascular areas (Fig. 12A,B,E). After phase II, the central avascular areas shrunk substantially in Phd2^f/f but only marginally in Phd2^EC−/− mice (Fig. 12C,D,F), suggesting that regenerative angiogenesis was defective in the latter. In terms of ischemic neovascularization, retinal areas occupied by neovascular tufts were drastically reduced in Phd2^EC−/− mice relative to Phd2^f/f mice (Fig. 12C,D,G). In addition to studies described in Fig. 12, we also carried out a parallel study using mice treated with tamoxifen at P0-P2, which led to the same conclusions (Fig. S7).

Because EC-specific deficiency of HIF2α (but not HIF1α) rescued retinal vascular development in Phd2^EC−/− background, we also analyzed {Phd2/Hif-1α}^EC−/− and {Phd2/Hif-2α}^EC−/− mice by the OIR model. At the end of phase I, the size of avascular areas was similar between these mice and their floxed controls (Fig. S8). At the end of phase II, {Phd2/Hif-1α}^EC−/− mice displayed phenotypes that were reminiscent of those in Phd2^EC−/− mice, including large avascular areas and scarcity of neovascular tufts (Fig. 13A,B, also compare Figs 12D and 13B). By contrast, phenotypes in {Phd2/Hif-2α}^EC−/− mice were similar to their floxed controls (Fig. 13C,D). These data indicate that EC-specific deletion of HIF2α but not HIF1α restored regenerative angiogenesis and ischemic neovascularization in Phd2^EC−/− background.

Because NOTCH suppression by DAPT affects both developmental retinal angiogenesis and ischemic retinal neovascularization (Dovey et al., 2001; Geling et al., 2002; Shi et al., 2013), we examined whether low dose DAPT could promote vascular growth in ischemic retinal tissues of Phd2^EC−/− mice. Briefly, Phd2^f/f and Phd2^EC−/− mice were exposed to 75% oxygen from P7 to P12 and then injected with DAPT (low dosage) or vehicle, twice each, first at P12 and then P14. At P17, whole-mount retinas were stained with IB₄, flat-mounted and analyzed by laser confocal imaging. In Phd2^f/f mice, DAPT had little effect (Fig. 14A,B,E,F). In Phd2^EC−/− mice, the DAPT-treated group had much smaller avascular areas than vehicle-treated controls (Fig. 14C-E). Unexpectedly, DAPT did not promote the formation of neovascular tufts in Phd2^EC−/− mice (Fig. 14C,D,F).

An additional noteworthy finding was that in the OIR model, DAPT promoted branching morphogenesis in peripheral retinal areas of Phd2^EC−/− mice, but it did not affect the Phd2^f/f mice (rectangular areas in Fig. 14A-D, with expanded views shown below each panel and quantification shown in Fig. 14G). Therefore, it appears that suppression of NOTCH signaling by low dose DAPT promoted regenerative angiogenesis both towards the avascular central area and within the peripheral retinal tissues, without driving the formation of neovascular tufts.

In this study, we investigated how retinal angiogenesis is regulated by oxygen sensing and NOTCH signaling in endothelial cells. Four features of the relevant phenotypes are worth highlighting. First, during normal retinal vascular morphogenesis, HIF1α and HIF2α proteins are present only at very low levels in retinal ECs, explaining why retinal angiogenesis is not diminished by EC-specific disruption of either or both of the Hif1alpha and Hif2alpha genes. Second, high level accumulation of HIF2α is responsible for diminished retinal angiogenesis in Phd2^EC−/− mice. Third, retinal vascular deficiency occurs in Phd2^EC−/− mice despite upregulated endothelial VEGFA levels. It is likely that upregulated VEGFA in retinal ECs disrupts the VEGFA gradient near the vascular front. Furthermore, VEGFA may upregulate DLL4, which is anti-angiogenic in the retina (Lobov et al., 2007). Fourth, developmental and regenerative retinal angiogenesis in Phd2^EC−/− mice can be significantly (although not completely) rescued by inhibiting NOTCH signaling, without simultaneously activating the formation of neovascular tufts.

The above observations invoke a previously unreported concept that HIF2α in retinal ECs is a suppressor to angiogenesis. An observation that appears to disagree with this notion is that EC-specific deletion of Hif2alpha alone does not cause excessive retinal angiogenesis. However, such a phenomenon can be partially explained by the fact that endothelial HIF2α levels are very low during normal retinal morphogenesis. EC-specific Hif2alpha deletion may not truly disturb normal retinal physiology because HIF2α levels are initially low. On the other hand, HIF2alpha accumulation would represent a deviation from the norm. This point is evident in Phd2^EC−/− mice, where retinal angiogenesis is severely suppressed by elevated HIF2α protein levels.

Our findings indicate that keeping endothelial HIF2α protein levels below a certain threshold constitutes a permissive condition to retinal angiogenesis. If endothelial HIF2α levels rise over such a threshold, as in the case of Phd2^EC−/− mice, the permissive state is compromised, in part due to upregulated DLL4. When this happens, angiogenic factors such as VEGFA are no longer permitted to efficiently induce angiogenesis, due to active NOTCH signaling in retinal ECs.

Besides in retinal ECs, DLL4 and HEY2 protein levels were also unexpectedly elevated in the INL/IPL regions of Phd2EC^−/− mice. Although a thorough investigation of this subject is beyond the scope of the current study, in future studies it may be pertinent to consider the following points. Do PHD2 deficient retinal ECs secrete a factor(s) that induce DLL4 expression in INL/IPL tissues? A recent preprint reported an example of DLL4 being secreted extracellularly (Wang et al., 2023 preprint). Is it possible that PHD2-deficient retinal ECs also secrete DLL4 in addition to producing membrane-tethered DLL4? If so, then diffusion of secreted DLL4 proteins to the INL/IPL region might cause local HEY2 upregulation through NOTCH signaling.

In considering why EC-specific Phd2 deletion led to reduced retinal angiogenesis, the spatial and temporal dynamics of retinal vascular morphogenesis is worth noting. After retinal angiogenesis is initiated at P0, the vascular front has more than 2 mm distance to travel before it reaches the retinal periphery (Duan et al., 2017). In the meantime, ECs immediately behind the VF quickly become oxygenated (West et al., 2005), a condition that favors PHD2-mediated HIF2α degradation. If HIF2α in retinal ECs were a required positive regulator, a gridlock situation would arise: new blood vessels would suppress their own growth and survival due to oxygen-dependent HIF2α degradation. If so, retinal angiogenesis would end soon after it is initiated. By permitting retinal angiogenesis to continue with declining abundance of endothelial HIF2α, the gridlock situation is avoided.

Findings from developmental studies were extended to the OIR model. Analyses in Phd2^EC−/− mice indicate that the formation of neovascular tufts may be suppressed by targeting Phd2 specifically in retinal ECs. It is also interesting to note that low dose DAPT led to the rescue of regenerative retinal angiogenesis without promoting ischemic neovascularization. It is not entirely clear why DAPT did not promote the formation of neovascular tufts, but the low dose applied in this study may be relevant. In a different study, delivery of higher dose of DAPT did promote the formation of neovascular tufts (Shi et al., 2013).

An anti-angiogenic role for endothelial HIF2α and NOTCH signaling may have the therapeutic potential to treat ocular diseases such as retinopathy of prematurity and diabetic retinopathy, both of which involve ischemic retinal neovascularization. Although additional work is needed to harness the clinical potential of this study, our data demonstrate the proof of principle that it is feasible to simultaneously suppress ischemic retinal neovascularization and promote regenerative angiogenesis by targeting both oxygen-sensing and NOTCH signaling mechanisms.

Housing and handling of mice were carried out according to animal protocols approved by Institutional Animal Care and Use Committee (IACUC) at UConn Health. Phd2^f/f and Hif2alpha^f/f mice were generated at UConn Health and have been described previously (Criscimanna et al., 2013; Takeda et al., 2006). Floxed Hif1alpha mice were purchased from the Jackson Laboratory (007561; originally donated by the Johnson lab, San Diego, CA, USA) (Ryan et al., 2000). The EC-specific Cdh5(Pac)Cre^ERT2 mice were originally generated in Ralf Adams' lab and have been used widely (Hao et al., 2017; Perrotta et al., 2022; Robson et al., 2010; Wang et al., 2010). Mice used in this study were in mixed background in C57BL/6 and CD1 at ∼75% to 25% ratio. A list of formal genotype nomenclature, availability, strain names and nicknames used in this article are provided in Table S1.

Individual neonatal mice (younger than P7) were identified by toe-clipping. Genomic DNA was extracted from clipped toe materials and used for genotyping by PCR. AmpliTaq DNA polymerase (Applied Biosciences) was activated at 95°C for 6 min. Each cycle consisted of 0.5 min at 94°C, 1 min at 58°C and 3 min at 72°C. Reactions were repeated for 35 cycles. PCR primer sequences and PCR band length are listed in Table S2.

All littermates, regardless the presence or absence of Cdh5(Pac)Cre^ERT2, were treated with tamoxifen by daily oral gavage from P0 through P2 (40 mg/kg per dose, dissolved in corn oil). EC specificity of tamoxifen-dependent Cdh5(Pac)Cre^ERT2 activation was confirmed by treating Ai9/Cdh5(Pac)Cre^ERT2 mice with tamoxifen, where Ai9 is a transgene expressing Cre-dependent TdTomato reporter driven by the ubiquitous CAG promoter. The Ai9 transgene is inserted into one of the two Rosa26 alleles (Fig. S2A-F) (Madisen et al., 2010), which are known to be dispensable for normal development and physiology (Friedrich and Soriano, 1991).

Retinas were isolated at P6. To obtain enough material, retinas from two or three mice were pooled and used as the starting material for EC isolation. Pooled retinas were minced into small pieces and incubated at 37°C for 30 min with type I collagenase (1 mg/ml in DMEM, serum-free, ThermoFisher, 17100017). Clumps were disrupted by pipetting up and down intermittently. Cells were pelleted by centrifugation (5 min, 1000 g). Supernatant was removed and cell pellets were resuspended in a fresh aliquot of collagenase solution to repeat the digestion. Cells were collected again by centrifugation, resuspended in 0.025% trypsin solution and DNase I (1 mg/ml), and incubated for 2 min at 37°C. Digestion was stopped by 10% FBS in DMEM. Cell suspensions were filtered through cell strainer (45 µm pore size, ThermoFisher). Cells passing through the filter were collected by centrifugation, resuspended in PBS containing 1% bovine serum albumin (BSA) and 1 mM EDTA, and incubated at 4°C for 1 h with PE anti-mouse CD31 (2 µg/ml, BioLegend, 102507) and FITC anti-mouse CD45 (2 µg/ml, BioLegend, 103107). Cell suspensions were applied to FACS ARIA II cell sorter and CD31⁺/CD45⁻ cells were directly sorted into TRIzol LS (ThermoFisher, 10296010).

Total RNA samples were prepared from retinal ECs directly sorted into TRIzol LS, reverse transcribed and used for qRT-PCR using primers listed in Table S3.

Eyes were enucleated from euthanized neonates and fixed in 4% paraformaldehyde (PFA) for 45 min at room temperature. Retinas were dissected from the fixed eyes under dissection microscopes. Each retina was partially cut to generate four similarly sized petals joined to one another through the optic nerve head (ONH).

Retinas were washed in phosphate-buffered saline (PBS) three times and then blocked in retina staining buffer (RSB, consisting of 1×PBS, 1 mM CaCl₂, 1 mM MgCl₂, 1% Triton X-100 and 1% BSA). Blocked retinas were incubated overnight at 4°C with primary antibodies (Table S4). After extensive washing, retinas were incubated with secondary antibody (Table S5) and Alexa Fluor 594-isolectin B₄ (IB₄) (2 µg/ml, ThermoFisher, l21413). In some cases, retinas were stained with only IB₄. In the study involving Ai9/Chd5(Pac)Cre^ERT2, Alexa Fluor 488-isolectin B₄ was used instead of Alexa Fluor 594-isolectin B₄ (Fig. S2). At completion of incubation, retinas were washed three times in PBS for 1 h each, and flat-mounted in 50% glycerol in PBS, supplemented with 1 mM CaCl₂ and 1 mM MgCl₂.

Whole eyes were enucleated from euthanized mice at P5 or P7 and fixed in 4% PFA for 45 min at room temperature. Fixed eyes were embedded in OCT, with the lens facing one of the vertical sides of mold. Sections were cut at 6 µm. Before staining, sections were washed in PBS and blocked and permeabilized in RSB. Sections were then incubated at 4°C for 2 h with primary antibodies (Table S4). After washing three times in PBS, sections were incubated with secondary antibody (Table S5) and Alexa Fluor 594-isolectin B₄ (IB₄).

Laser confocal imaging was carried out with Zeiss LSM 880 microscope at UConn Health CCAM (Imaging Core). Images were recorded with ZEN software (Zeiss). Owing to large specimen sizes of retinas, images were often montaged from individual panels. Raw images were further cropped and labeled in Photoshop and/or Adobe Illustrator.

In each retina, quantification was carried out in two of the four petals. The average of a total of four values from both retinas was used as one data point in statistical analysis. To determine microvascular density, image pixel density was set at 300 pixels per inch (ppi) and image dimension was set at 254×50.8 mm² for a physical specimen area of 1×0.2 mm². For junctional (branching) points, image pixel density was at 72 ppi, and image dimension at 92×18.4 mm² for a 1×0.2 mm² physical specimen area. Measurements were carried out with AngioTool (Zudaire et al., 2011). Additional settings within AngioTool were as follows: intensity, 21; ‘remove small particles’, 8; ‘fill holes’, default. The distance between the VF and ONH was determined with the aid of NIH ImageJ.

Whole-mount retinas were stained using IB₄ and using anti-EGR, flat-mounted and analyzed by laser confocal imaging. Z-stack confocal images (8 µm) were generated using Imaris software (Bitplane). Normal tip cells were identified as cells with the following features: (1) IB₄⁺; (2) located at the vascular front with length of the cell body protruding towards avascular area; (3) containing large ERG⁺ nuclei; and (4) displaying extended filopodia. Abnormal tip cells were identified similarly, except that filopodia were either lacking or blunted. The number of normal and abnormal tip cells was quantified with NIH ImageJ.

For the proliferation assay, mice were injected intraperitoneally with 5′-bromo-2′-deoxyuridine (BrdU, 120 mg/kg) at P6 and euthanized 1 h later. Retinas were isolated from enucleated eyes, fixed in 4% PFA (45°C, room temperature) and treated in 2 mM hydrochloric acid (HCl) for 1 h at 37°C. Retinas were then blocked in RSB and double-stained with IB₄ and biotin-labeled anti-BrdU (Table S4), the latter of which was detected by Streptavidin-Alexa Fluor-488 (Table S5). The number of BrdU⁺ cells in a specified area was counted with the aid of NIH ImageJ; the combined length of vascular fragments in the same area was measured using AngioTool.

Apoptotic ECs were detected by two independent methods, including immunofluorescence staining with anti-CAS3 (active fragment) (whole mount) and TUNEL staining (cryosections). All specimens were also stained with IB₄ to label blood vessels. After IB₄ and anti-CAS3 staining, retinas were flat-mounted and analyzed by confocal imaging. Active CAS3 levels were quantified as the ratio between CAS3⁺ and IB₄⁺ pixel values. TUNEL staining was carried out with an In Situ Cell Death Detection Kit according to the manufacturer's instructions (Roche, 11684795910). Briefly, eyes were isolated at P5 and fixed with 4% PFA, and cryosections were cut at 6 µm. Cryosections were permeabilized with 0.1% Triton X100 and incubated with TUNEL reaction mix (a cocktail including buffer, dNTPs, Fluorescein-dUTP and Terminal Deoxynucleotidyl Transferase). After 60 min at 37°C, sections were washed three times in PBS, and further incubated with IB₄ overnight at 4°C. Stained sections were analyzed by confocal microscopy.

DAPT [N-{N-(3,5-DifluorophenacetylL-alanyl)}-S-phenylglycine t-butylester, Sigma-Aldrich, 565770] was dissolved at a final concentration of 37.5 mg/ml in peanut oil containing 10% ethanol. DAPT was injected intraperitoneally to neonatal mice essentially according to a published protocol, but at a substantially lower dose and frequency (Benedito et al., 2009). Instead of daily injection at 200 mg/kg per dose, DAPT injection was carried out every other day at 50 mg/kg per dose, for a total of two doses (P3 and P5 in developmental studies; P12 then P14 in the OIR model).

Anti-DLL4 (Table S4) was injected into the vitreous cavity of Phd2^EC−/− mice at P4 (1 µl per eye and 1 µg/µl in PBS). In separate Phd2^EC−/− mice, hamster IgG was injected in the same way as anti-DLL as a control. Injection was carried out using a gastight Hamilton syringe outfitted with a 33-gauge needle.

Fluorescence signals on confocal images were measured as pixel values per specified area with the aid of NIH ImageJ. Accuracy and reliability were assured by the combination of two methods. First, the same microscope and camera settings were applied to different samples of the same analysis; second, background pixel values were measured in apparently blank areas and subtracted from the raw pixel values.

For mice used in the oxygen-induced retinopathy (OIR) model, tamoxifen treatment was carried out at P4-P6 (40 mg/kg per dose) (Fig. S1F, Figs 12 and 13), except for the study shown in Fig. S7, in which tamoxifen was delivered at P0-P2. The OIR model was set up as described previously (Pierce et al., 1995). In phase I, neonatal mice were placed in 75% oxygen chamber from P7 to P12. Subsets of mice were euthanized at the end of Phase I to quantify the extent of microvascular obliteration. To start phase II, mice were returned to ambient room air immediately after phase I. At the end of phase II (P17), mice were euthanized, and retinas were isolated and stained with IB₄. Stained retinas were flat-mounted and imaged by confocal microscopy. Total and avascular retinal areas were measured with NIH ImageJ. Retinal areas occupied by neovascular tufts were measured with the magic wand tool in Photoshop (Connor et al., 2009).

Sample sizes (n) are the number of mice per group, or the number of pools per group in studies involving cell sorting and qRT-PCR (retinas from two or three mice of the same genotype per pool). Data were analyzed by two-tailed Student's t-test. P<0.05 were considered statistically significant. Variations are standard deviation (s.d.). All statistical analyses and graph plotting were carried out in Excel.

Confocal images were taken at the University of Connecticut Health Center (UConn Health) CCAM Microscopy Facility (Imaging Core) on a fee per 30 min basis. Cell sorting was carried out with the help of UConn Health Flow Cytometry Core on a fee per full-service basis. Cryosections were cut at the UConn Health Research Histological Core.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202802.reviewer-comments.pdf