Vascular pruning is crucial for normal development, but its underlying mechanisms are poorly understood. Here, we report that retinal vascular pruning is controlled by the oxygen-sensing mechanism in local astrocytes. Oxygen sensing is mediated by prolyl hydroxylase domain proteins (PHDs), which use O2 as a substrate to hydroxylate specific prolyl residues on hypoxia inducible factor (HIF)-α proteins, labeling them for polyubiquitylation and proteasomal degradation. In neonatal mice, astrocytic PHD2 deficiency led to elevated HIF-2α protein levels, expanded retinal astrocyte population and defective vascular pruning. Although astrocytic VEGF-A was also increased, anti-VEGF failed to rescue vascular pruning. However, stimulation of retinal astrocytic growth by intravitreal delivery of PDGF-A was sufficient to block retinal vascular pruning in wild-type mice. We propose that in normal development, oxygen from nascent retinal vasculature triggers PHD2-dependent HIF-2α degradation in nearby astrocytic precursors, thus limiting their further growth by driving them to differentiate into non-proliferative mature astrocytes. The physiological limit of retinal capillary density may be set by astrocytes available to support their survival, with excess capillaries destined for regression.
Retinal vascular development is spatially and temporally coordinated with the development of the astrocytic network (Dorrell et al., 2002; Duan et al., 2017; Gariano et al., 1996; Ling et al., 1989). In rodents, retinal astrocytic development begins as early as the late-gestation stages, with astrocytic progenitor cells (APCs) migrating out of the rim of the optic nerve head (ONH) and onto the inner retinal surface (Ling et al., 1989). APCs are strongly Pax2+ in their nuclei but negative for glial fibrillary acidic protein (GFAP) (Chan-Ling et al., 2009; Duan et al., 2017). At birth, APCs have spread out to approximately halfway between the ONH and retinal periphery (Duan et al., 2017; Huxlin et al., 1992; Ling et al., 1989), but mostly remain in the progenitor stage. Once retinal angiogenesis begins in newborn mice, APCs undergo partial differentiation to become immature astrocytes (IACs) ahead of the vascular front, and full differentiation to become mature astrocytes (mASCs) within the vascularized area. IACs remain strongly Pax2+as in APCs, but distinguish themselves from APCs by gaining weak GFAP expression. By contrast, mASCs express lower levels of Pax2, but are strongly GFAP+ (Duan et al., 2017). APCs and IACs are astrocytic precursors at different stages of differentiation towards full maturity. Astrocytic precursors are proliferative, but mASCs are generally quiescent. Furthermore, both APCs and IACs are highly migratory, with IACs approaching the retinal periphery by postnatal day (P) 3 in mice (Dorrell et al., 2002; Duan et al., 2017).
In rodents, retinal angiogenesis begins at birth, with nascent blood vessels sprouting out of the rim of the ONH (Fruttiger, 2002). In the subsequent week, the vascular front continues to advance in radial directions towards the retinal periphery, led by vascular tip cells (Fruttiger, 2002; Gerhardt et al., 2003; Scott et al., 2010). Whereas retinal angiogenesis is first induced by APCs after birth, subsequent vascular growth towards the retinal periphery depends on IACs. As they form, new blood vessels provide oxygen to nearby IACs, driving them to differentiate into mASCs (Duan et al., 2017). In addition to oxygen, endothelial cell (EC)-derived factors, such as leukemia inhibitory factor and the apelin/APJ system, also promote retinal astrocytic differentiation (Fukushima et al., 2009; Sakimoto et al., 2012).
Oxygen is a substrate of prolyl hydroxylase domain (PHD) proteins (also referred to as EGLNs), a subfamily of oxoglutarate and Fe2+-dependent dioxygenases that hydroxylate specific prolyl residues in hypoxia inducible factor (HIF)-α (Bruick and McKnight, 2001; Epstein et al., 2001; Ivan et al., 2002). Once hydroxylated, HIF-α proteins undergo pVHL (VHL)- and E3 ligase-dependent polyubiquitylation, followed by degradation in proteasomes (Epstein et al., 2001; Huang et al., 2002; Maxwell et al., 1999). Under normal physiological conditions, HIF-α protein levels are tightly controlled by tissue oxygen levels. Rising oxygen levels due to vascular development may activate PHD2 (EGLN1), triggering HIF-α hydroxylation and degradation. In neonatal mouse retinas, expression of HIF-2α (EPAS1) in astrocytic precursors (APCs and IACs) is crucial for the maintenance of their proliferative capacity (Duan et al., 2014), whereas diminished HIF-2α levels may cause APCs and IACs to differentiate into proliferation-incompetent mASCs (Duan et al., 2017, 2014).
The development of the primary retinal vasculature proceeds in two phases: (1) formation of nascent vascular plexus by sprouting angiogenesis; and (2) remodeling of the honeycomb-patterned nascent vascular plexus into a mature vascular tree displaying hierarchical organization of larger trunks and smaller branches (Fruttiger, 2007). Although angiogenesis has been studied extensively (Bentley et al., 2014; Costa et al., 2016; Hellström et al., 2007; Jakobsson et al., 2010; Pitulescu et al., 2017; Sawamiphak et al., 2010; Tamagnone and Mazzone, 2011; Tammela et al., 2011), much less is known about developmental remodeling. Nonetheless, it is well recognized that developmental remodeling involves multiple regulated processes, including recruitment of mural cells, regulation of vascular calibers, and pruning (regression) of excess microvessels (Arreola et al., 2018; Cheng et al., 2012; Chu et al., 2016; Ehling et al., 2013; Ishida et al., 2003; Korn et al., 2014; Phng et al., 2009; Scott et al., 2010; Simonavicius et al., 2012; Watson et al., 2016; Zaitoun et al., 2015).
In this study, we demonstrate that oxygen sensing in retinal astrocytes controls endothelial growth and apoptosis, morphologically manifested as regulation of vascular pruning and capillary caliber. Blocking oxygen sensing in astrocytes by targeted disruption of the Phd2 locus caused HIF-2α accumulation, lingering of astrocytes in immature states, expansion of the retinal astrocyte population, and defective vascular pruning. These findings indicate that oxygen targets retinal astrocytes instead of endothelial cells to initiate developmental vascular regression.
Disruption of Phd2 in retinal astrocytes
We disrupted the Phd2 gene in astrocytes to cripple their oxygen-sensing mechanisms. Although several different GFAPCre mouse lines are available to delete floxed genes in astrocytes, only one of them (Jackson Laboratory stock number 004600, originating from the Messing Lab, University of Wisconsin, WI, USA) has demonstrated Cre activity in the precursors of retinal astrocytes (Duan et al., 2014; Zhuo et al., 2001). Thus, the Jackson Laboratory 004600 GFAPCre line was chosen for this project to achieve Phd2 disruption early in retinal astrocyte development.
To assess Phd2 targeting efficiency, we quantified PHD2 mRNA and protein levels in neonatal retinal astrocytes cultured according to an established protocol, with minor modifications (Fig. S1) (Scheef et al., 2005). According to both quantitative RT-PCR (qPCR) and western blotting, astrocytic PHD2 level in Phd2f/f/GFAPCre mice was reduced to 6% or less relative to floxed controls (Fig. 1). qPCR and ELISA also revealed that VEGF-A expression was increased in PHD2-deficient retinal astrocytes (Fig. 1A,B), but EPO expression was not significantly affected. In addition, PDGF-A expression was undetectable in both floxed and PHD2-deficient astrocytes. Western blotting demonstrated that HIF-1α and HIF-2α were increased by 1.6- and 1.8-fold, respectively, in Phd2f/f/GFAPCre mice compared with controls (Fig. 1C). However, HIF-1α bands remained relatively weak even after a 1.6-fold upregulation owing to a very low basal level. Interestingly, astrocytic HIF-2α but not HIF-1α was previously found to be crucial for retinal astrocytic and vascular development in neonatal mice (Duan et al., 2014).
Astrocytic PHD2 deficiency causes retinal astrocytes to linger in less mature states
Retinal astrocytic differentiation was examined by whole-mount anti-GFAP and anti-Pax2 immunofluorescence (IF) staining, followed by laser confocal imaging of flat-mount retinas (Fig. 2A-J). In normal retinal development, a boundary is typically discernable between the strongly GFAP+ region on the central side and weakly GFAP+ region on the peripheral side. This boundary represents the front edge of the mature astrocyte cell population, and is referred to as the mASC front (Duan et al., 2017). At P3, the mASC front was less advanced towards the retinal periphery in Phd2f/f/GFAPCre mice than in floxed controls (Fig. 2A,B,M). At P5 and P8, however, the mASC front was at similar positions in floxed and Phd2f/f/GFAPCre mice (Fig. 2C-F,M).
Behind the mASC front, strongly GFAP+ structures organized into hierarchical patterns in floxed mice but appeared as a random collection of branches in Phd2f/f/GFAPCre mice. The edges of GFAP+ structures were typically fuzzy in Phd2f/f/GFAPCre mice, appearing as though they were out of focus (Fig. 2A-F, insets). In addition, large GFAP+ structures in Phd2f/f/GFAPCre mice often appeared dimmer than in their floxed controls. These features partially resembled those in the peripheral side of the mASC front in floxed mice, implying that the retinal astrocytes might be less mature in Phd2f/f/GFAPCre mice.
As a first step towards analyzing retinal astrocyte differentiation in Phd2f/f/GFAPCre mice, we quantified the GFAP+ signals as a percentage of GFAP+ pixels per area (% GFAP+ area) (Fig. 2N). Areas selected for quantification were approximately halfway between the ONH and the mASC front in Phd2f/f mice, or at a comparable distance to the ONH in Phd2f/f/GFAPCre mice (Fig. 2A,C,E). As summarized in Fig. 2N, Phd2f/f/GFAPCre mice had lower % GFAP+ area at P3, but caught up with Phd2f/f mice by P5, and eventually surpassed Phd2f/f mice by P8. Of note, % GFAP+ area values should be interpreted with caution, because they are influenced not only by the differentiation status of astrocytes but also their population density. For example, a very crowded population of weakly GFAP+ immature astrocytes could potentially lead to high % GFAP+ area values because of more numerous cells in the region. Thus, proper assessment of retinal astrocyte differentiation would require normalization of GFAP+ signals to astrocyte abundance.
Pax2 is an appropriate marker for normalization because it is expressed in all retinal astrocytes, with expression gradually declining as differentiation progresses to more mature states. However, the Pax2 expression pattern itself is also informative of astrocyte distribution in different retinal regions. Therefore, we first characterized the retinal distribution of Pax2+ staining signals in floxed and Phd2f/f/GFAPCre mice. At P3, Pax2 expression patterns appeared grossly similar between floxed and Phd2f/f/GFAPCre mice (Fig. 2G,H). However, closer inspection indicated that in Phd2f/f/GFAPCre mice, Pax2+ cells were more densely packed near the ONH (Fig. 2G,H). This observation was confirmed by quantification of % Pax2+ area (Fig. 2O).
Next, we quantified retinal astrocyte differentiation by normalizing % GFAP+ area to % Pax2+ area in double-stained retinas. A higher GFAP/Pax2 ratio indicates more advanced astrocyte differentiation, whereas a reduced ratio suggests less mature status (Duan et al., 2017). Phd2f/f/GFAPCre mice had reduced GFAP/Pax2 ratios at all three stages examined (P3, P5 and P8) (Fig. 2P). Thus, retinal astrocytes in Phd2f/f/GFAPCre mice were at less mature states than in floxed mice. Also, although Phd2f/f/GFAPCre mice displayed modestly elevated % GFAP+ area values at P8, this feature by itself did not indicate more advanced astrocyte differentiation. Instead, because the GFAP/Pax2 ratio was lower, we can conclude that retinal astrocytes were also less mature in P8 Phd2f/f/GFAPCre mice. The range of GFAP/Pax2 ratio in Phd2f/f/GFAPCre is between 1 and 2. According to the criteria defined in a previous study (Duan et al., 2017), these values indicate intermediate differentiation status. Overall, these data demonstrate that maturation of retinal astrocytes is hindered in Phd2f/f/GFAPCre mice.
As immature astrocytes are proliferative whereas mature astrocytes are quiescent (Nahirnyj et al., 2013), the less mature status of retinal astrocytes in Phd2f/f/GFAPCre mice might lead to an expanded astrocyte population. In support of this notion, the number of Pax2+ nuclei was indeed increased in PHD2-deficient mice (Fig. 2Q). Retinal astrocyte abundance may be also quantified as % PDGFRα+ area in anti-PDGFRα-stained retinas. Because PDGFRα is expressed in immature as well as mature astrocytes (Fruttiger, 2002; Mudhar et al., 1993), % PDGFRα+ area values are inclusive of astrocytes at different stages of differentiation. As shown in Fig. 2K,L,R, % PDGFRα+ area values were higher in Phd2f/f/GFAPCre mice, demonstrating that retinal astrocyte abundance was increased.
Astrocytic PHD2 deficiency leads to defective APC migration in a small percentage of mice
The phenotypes presented above were observed in 91% of Phd2f/f/GFAPCre mice, and are referred to as Phenotype A. The remaining 9% of Phd2f/f/GFAPCre mice displayed Phenotype B, in which Pax2+ cells failed to migrate towards the retinal periphery (Fig. S2). Also, astrocyte differentiation was mostly limited to the immediate vicinity of ONH, whereas the majority of Pax2+ cells beyond this region remained GFAP− as late as P5, indicating that they remained at the APC stage (Fig. S2F, area 1). In normal development, APCs are abundantly present only in newborn mice, but quickly become IACs or mASCs by P1, depending on their relative location with the developing vasculature (Duan et al., 2017). Although failed APC migration was unique to Phenotype B, persistence of APCs was probably a more severe form of differentiation defects seen in Phenotype A.
Because mice used for these studies were in a mixed strain background, we wondered whether backcrossing to C57BL/6 would result in improved consistency. In mice with 97% C57BL/6 background, Phenotype B was not observed. Perplexingly, Phenotype B also disappeared in mice with 95% CD1 background. Although these outcomes did not allow us to identify the exact strain background responsible for Phenotype B, they nevertheless indicated that the existence of two different phenotypes was related to a peculiar mix of strain background. Given that Phenotype A was relatively insensitive to strain background (i.e. the penetrance was at least 91% in outbred background), we mostly focused on Phenotype A, while also performing some preliminary characterization of Phenotype B.
Astrocytic PHD2 deficiency leads to vascular overcrowding
In Phd2f/f/GFAPCre mice, a major vascular defect at P3 and P5 was the overcrowding of endothelial cells near the ONH, with the central retina being uniformly IB4+ (Fig. S3A-D). Further away from the ONH, individual blood vessels were discernable but failed to organize into a hierarchical pattern. Retinal blood vessels mostly had similar sizes instead of remodeling into large vascular trunks and smaller branches. Vascular density was apparently higher than in floxed controls, except in the area just behind the vascular front. Consistent with these qualitative observations, % IB4+ area in the central retina was significantly higher in Phd2f/f/GFAPCre mice (Fig. S3G). However, the narrow area immediately behind the vascular front appeared similar to floxed mice.
In mice displaying Phenotype B, ECs clumped together near the ONH but failed to migrate towards the retinal periphery (Fig. S3E,F). Because EC migration during retinal vascular development is normally dependent on immature astrocytes (IACs) ahead of the vascular front, failed EC migration in Phenotype B appeared to be a secondary defect to the lack of IACs.
For more detailed analysis of the vascular defects, we focused on Phenotype A. At P8, the retinal vasculature had undergone extensive developmental remodeling in floxed mice but not in Phd2f/f/GFAPCre mice (Fig. 3A,B). Similar to earlier stages (P3 and P5), Phd2f/f/GFAPCre mice had higher % IB4+ area values in most retinal areas, except just behind the vascular front (Fig. 3A,B,G). In addition to % IB4+ area, we also quantified the number of vascular branches in mid-retinal areas (i.e. halfway between the ONH and retinal periphery). Within this general area, three different sub-areas were quantified separately. At all three locations, Phd2f/f/GFAPCre mice had more numerous vascular branches than did floxed mice, but this was most pronounced near arterioles (Fig. 3C,D,H).
Microvascular diameter was measured at two retinal locations: near the vascular front and in the mid-retinal area. To minimize the inherent inaccuracy of measuring small capillary sizes, we first measured the sum of diameters for each set of six capillaries as shown in Fig. 3E,F. All six capillaries were from the same specified retinal area, and the average of each set from the same mouse was used as one data point for statistical analysis. The results indicated that in mid-retinal areas, capillaries in Phd2f/f/GFAPCre mice were larger than those in the floxed group (Fig. 3I). Just behind the vascular front, the two groups had similarly large capillaries, which was not surprising because in normal mice the capillaries near the vascular front are immature and relatively large.
Because astrocyte abundance and vascular density were both increased in Phd2f/f/GFAPCre mice (Phenotype A), we wondered whether the two phenotypes were related to each other. Therefore, we directly compared astrocyte abundance and capillary density in P5 retinas double stained with IB4 and anti-Pax2. In Phd2f/f/GFAPCre mice, higher capillary density matched the increased number of Pax2+ cells at the same retinal location (Fig. 4). Because Phd2 disruption was achieved with astrocyte-specific Cre, we propose that astrocytic phenotypes caused the vascular phenotypes.
Astrocytic PHD2 deficiency promotes EC proliferation but suppresses tip cell filopodia formation
We examined EC proliferation by measuring 5′-bromo-2′-deoxyuridine (BrdU) incorporation (Fig. 5). Quantification of IB4+/BrdU+ cells was carried out at four retinal locations, indicated by rectangles a, b, c and d (Fig. 5A,B). Just behind the vascular front (a), IB4+/BrdU+ cells were abundant in both Phd2f/f and Phd2f/f/GFAPCre mice. Further behind (b, c and d), IB4+/BrdU+ cells were significantly more abundant in Phd2f/f/GFAPCre mice (Fig. 5A-C). These observations indicate that astrocytic PHD2 deficiency led to persistent EC proliferation in areas where they were normally quiescent, thus contributing to increased capillary caliber.
To monitor sprouting angiogenesis directly, we quantified the number of vascular tip cells (Fig. 6). Phd2f/f/GFAPCre mice had normal numbers of tip cells (Fig. 6A,B,E), but there were fewer filopodia per tip cell (Fig. 6C-E). Nonetheless, microvascular density immediately behind the tip cells was similar in Phd2f/f and Phd2f/f/GFAPCre mice. Because pruning and other remodeling events in this nascent vascular area is negligible, vascular abundance just behind the tip cells mostly reflect the activity of angiogenesis. Thus, our data suggest that sprouting angiogenesis was essentially normal in Phd2f/f/GFAPCre mice displaying Phenotype A.
Astrocytic PHD2 deficiency suppresses EC apoptosis and vascular regression
EC apoptosis was examined by double staining of whole-mount retinas with IB4-Alexa 594 and anti-caspase 3 (directed to the active fragment, Cas3) (Fig. 7). To validate Cas3 whole-mount IF staining, we tested the antibody in positive controls generated by overnight exposure of P7 mice to 75% O2. This pilot test demonstrated the presence of numerous IB4+/Cas3+ in the retinas (Fig. 7A,D,G,J). The effect of astrocytic PHD2 deficiency on EC apoptosis was assessed in P6 mice maintained under normoxia at all times. In the Phd2f/f group, seven IB4+/Cas3+ apoptotic ECs were detected per 1.0 mm2 retinal area, but such cells were virtually undetectable in the Phd2f/f/GFAPCre group (<1 per mm2) (Fig. 7B,C,E,F,H,I,K-M). These findings indicate that astrocytic PHD2 deficiency blocked retinal EC apoptosis.
In an alternative method of monitoring microvascular regression, we visualized remnant microvascular fragments by double staining with IB4-Alexa 594 and anti-collagen IV (Fig. 8A-F). Capillary regression transiently leaves behind remnants of their original vascular components, such as collagen in the basement membrane, fragments of EC cells (IB4+), or a mixture of both. The exact content and relative proportions of these remnant structures may depend on which component degrades faster in a particular microvascular environment. To simplify the analysis, we grouped regressing microvessels into three categories: (1) collagen IV+/IB4−; (2) collagen IV+/IB4+; and (3) collagen IV−/IB4+. Examples for each are illustrated in Fig. 8G,H. In both floxed and PHD2-deficient mice, regressing microvessels were infrequent near the vascular front (Fig. 8C,D). Further behind, all three categories were abundantly present in floxed mice, but rare in Phd2f/f/GFAPCre mice (Fig. 8E,F,I). Taken together, our data demonstrate that astrocytic PHD2 deficiency blocked developmental vascular regression.
Increased retinal astrocyte population density, but not VEGF-A or T-cell abnormality, blocks vascular pruning
In exploring the potential mechanisms underlying blocked retinal vascular pruning in Phd2f/f/GFAPCre mice, we first considered whether increased VEGF-A expression might be responsible for the blockade of capillary regression. To examine this possibility, we chose to perform intravitreal injection of anti-VEGF. Because overdosage of anti-VEGF might block sprouting angiogenesis altogether, which would preclude the analysis of vascular pruning, pilot experiments were performed to optimize anti-VEGF dosage. Subsequently, we injected P4 Phd2f/f/GFAPCre mice with 1.8 µg of anti-VEGF per eye, or PBS in the contralateral eye. When examined at P8, retinas from anti-VEGF-injected eyes displayed partially inhibited sprouting angiogenesis, confirming that VEGF-A was effectively neutralized (Fig. 9). However, vascular pruning remained blocked. These data indicate that although astrocytic VEGF-A expression was elevated in Phd2f/f/GFAPCre mice, it may not be responsible for blocked vascular pruning.
Next, we investigated whether failed vascular pruning in Phd2f/f/GFAPCre mice was related to any T-cell abnormality. This possibility was considered because of an earlier report suggesting that apoptotic signals from T cells were essential for EC apoptosis and retinal vascular pruning during development (Ishida et al., 2003). Because there is no known connection between oxygen sensing and T-cell function, we first sought to confirm that T cells were indeed necessary for retinal vascular pruning. Therefore, we examined retinal vascular development in SCID (severe combined immunodeficiency) mice. Unexpectedly, retinal vascular patterning was normal in neonatal SCID mice (Fig. S4). Therefore, at least in our hands, T cells did not appear to be important for retinal vascular pruning. Given this outcome, no further studies were carried out on T cells.
Finally, we turned our attention to the intriguing possibility of excessive astrocytic growth as a cause of failed vascular pruning. If astrocyte overgrowth was the underlying mechanism, then it should be possible to block vascular pruning in wild-type mice by stimulating sustained astrocyte growth without directly manipulating the oxygen-sensing system. Because PDGF-A is known to promote the proliferation and migration of immature astrocytes (Nakamura-Ishizu et al., 2012), we performed intravitreal injection of PDGF-A in Phd2f/f mice. Injection was performed at P4, and retinas were examined at P8. For confocal imaging analysis, we focused on the peripheral half of the retina because that was where vascular development occurred after injection (Fig. 10).
PDGF-A effectively blocked retinal vascular pruning in Phd2f/f mice (Fig. 10A-D,I). Microvascular caliber was also larger in PDGF-A-injected mice (Fig. 10C,D,J). Given that PDGF-A is not known to directly promote retinal EC growth or survival, we suspected that its effects on the retinal vasculature might be due to increased astrocyte growth. Indeed, retinas from PDGF-A-injected eyes exhibited increased astrocyte abundance (Fig. 10E-H,K,L). More importantly, the GFAP/Pax2 ratio was reduced, suggesting that PDGF-A maintained the astrocytes in less mature states thus permitting their continued growth (Fig. 10M).
Astrocytes, oxygen, and vascular development
The earliest hint that retinal astrocytes had a role in vascular development came from the observation that the avascular retinas in possums were also free of astrocytes (Stone and Dreher, 1987). Subsequent studies indicated that hypoxia in retinal astrocytes may contribute to their proangiogenic functions. For example, Nr2e1 (nuclear receptor subfamily 2 group E member 1, a homolog of the Drosophila tailless gene Tll) is much more strongly expressed in immature retinal astrocytes in the hypoxic avascular region than in the mature astrocytes in vascularized retinal tissues. Knockout of Nr2e1/Tlx almost completely blocked retinal angiogenesis, implying that its proangiogenic role could be also hypoxia-related, in addition to its role in regulating fibronectin deposition (Miyawaki et al., 2004; Uemura et al., 2006). A more direct relationship between hypoxia in retinal astrocytes and angiogenesis was demonstrated in mice lacking astrocytic HIF-2α, where the astrocytic network was poorly developed and retinal angiogenesis significantly suppressed (Duan et al., 2014).
Oxygen is known to suppress retinal vascular development or cause existing microvessels to regress. For example, exposure of newborn mice to 75% oxygen completely blocked retinal vascular development (West et al., 2005). In the mouse model of oxygen-induced retinopathy, exposure of neonatal mice to hyperoxia between P7 and P12 leads to massive obliteration of microvessels in the central retina, but the relevance of this disease model to developmental vascular pruning is uncertain (Connor et al., 2009; Smith et al., 1994). A potential role of oxygen in developmental vascular pruning is implied by the presence of avascular zones near arterioles, but its mechanism of action is unclear. Phd2f/f/GFAPCre mice generated in the present study provided a valuable tool for investigating developmental vascular pruning, and allowed us to identify retinal astrocytes as the major targets of oxygen to regulate vascular pruning.
Mechanisms of retinal vascular pruning
Because hypoxia upregulates VEGF-A expression, one might speculate that astrocytic PHD2 deficiency, which molecularly mimics hypoxia, blocks retinal vascular pruning by upregulating VEGF-A expression in retinal astrocytes. Although VEGF-A was indeed upregulated, anti-VEGF failed to rescue vascular pruning. Furthermore, it has been shown previously that astrocyte-specific VEGF-A disruption led to very minimal changes in retinal vascular patterning (Scott et al., 2010). Therefore, VEGF-A regulation does not appear to be directly involved in vascular pruning.
Our data suggest that the oxygen-sensing mechanism controls vascular pruning by regulating astrocyte abundance. Oxygen from the nascent vascular plexus may activate PHD2 in nearby astrocytes, leading to the downregulation of astrocytic HIF-2α protein levels. Note that astrocytes in newly vascularized areas are relatively immature when they are first exposed to blood vessels, but HIF-2α downregulation in response to rising oxygenation levels may cause them to quickly differentiate into mature astrocytes. This view is consistent with our earlier finding that astrocytic HIF-2α deficiency greatly accelerated retinal astrocyte differentiation towards the mature state (Duan et al., 2014). Data presented in this paper complement our previous finding in astrocyte-specific HIF-2α-deficient mice and demonstrate that astrocytic PHD2 deficiency hinders retinal astrocyte maturation, likely through HIF-2α upregulation. Because astrocytes are more proliferative in less mature states, retinal astrocyte growth likely continued for longer time periods in Phd2f/f/GFAPCre mice, thus leading to significantly expanded astrocyte populations.
Because blood vessels in the primary retinal vasculature are generally supported by adjacent astrocytes, we propose that the physiological limit of vascular density is set by local astrocyte abundance. In other words, only those blood vessels with adequate astrocytic support may survive. Capillaries in excess of available astrocytes are destined to regress (Fig. 11). We confirmed the pivotal role of retinal astrocyte abundance in determining capillary density by an independent approach. Intravitreal delivery of PDGF-A, which stimulates astrocyte growth through its receptor signaling but is not known to modulate the oxygen-sensing mechanism, not only increased astrocyte abundance but also blocked vascular pruning.
Dependence of APC migration on the oxygen-sensing mechanism
Besides regulating astrocyte differentiation and vascular pruning, astrocytic PHD2 also seems to participate in sensing the oxygen gradient. During normal development of the primary retinal vasculature, an oxygen gradient may exist across the retina, with the vascularized region being well oxygenated and the more peripheral avascular region being hypoxic (Gerhardt et al., 2003). As PHD2 uses O2 as a substrate, it follows that its hydroxylase activity in astrocytes may be dictated by local oxygen concentration, decreasing in the direction of the ONH towards the retinal periphery. In Phd2f/f/GFAPCre mice, astrocytes and their precursors are desensitized to oxygen owing to PHD2 deficiency. Therefore, the accumulation of APCs near the ONH in Phenotype B mice suggests that their migration depends on PHD2-mediated oxygen sensing. It is unclear why the migration defect is present only in a low percentage of Phd2f/f/GFAPCre mice, but future studies should examine whether double or triple knockout of different PHD isoforms might increase the penetrance of this phenotype.
In summary, developmental vascular pruning may be viewed as a mutual feedback mechanism between the nascent microvascular plexus and the associated astrocytes. Rising oxygen levels accompanying retinal angiogenesis may limit the expansion of the astrocytic population by driving maturation differentiation, thus forcing excess microvessels without astrocytic support to undergo regression. Given that astrocyte abundance is controlled by vascularization via astrocytic oxygen-sensing mechanism, it can be said that vascular pruning is ultimately auto-regulated by the vascularization process itself.
MATERIALS AND METHODS
All animals were handled according to procedures approved by the Institutional Animal Care and Use Committee (IACUC) at the UConn Health in compliance with Animal Welfare Assurance. Floxed Phd2 mice were originally generated in our lab, using C57BL/6/129 S6 hybrid embryonic stem cells targeted at the Phd2 locus (Takeda et al., 2006). GFAPCre transgenic mice were purchased from the Jackson laboratory (stock number 004600), and were in FVB strain background upon arrival. Phd2f/f/GFAPCre mice used were in mixed background of CD1/C57BL/6/129 S6 /FVB, at estimated relative proportion of 0.13:0.56:0.06:0.25. SCID mice were purchased from the Jackson Laboratory (stock number 001303).
Ear punch tissues from weaned mice were used for DNA extraction and PCR. For neonatal mice (P8 or younger), toe clipping was performed to identify individual mice and provide tissues for genotyping. PCR primers are summarized in Table S1.
Primary cultures of neonatal retinal astrocytes
Retinas were isolated from neonatal mice at P3, and minced with fine dissection scissors. The retinal pieces were incubated in digestion medium for 45 min at 37°C, with gentle rocking and intermittent pipetting up and down to further break the pieces. Digestion medium was serum-free DMEM (pre-equilibrated in 5% CO2) supplemented with Collagenase Type I (1 mg/ml, Life Technologies) and DNase I (100 µg/ml, Life Technologies). After incubation, cell suspensions were mixed with equal volume of pre-warmed (37°C) 0.05% trypsin (Life Technologies), and incubated for another 2 min. Trypsin digestion was terminated by the addition of fetal bovine serum (FBS) to 10%. Cells were centrifuged at 1000 g for 5 min, re-suspended in DMEM containing 10% FBS, and filtered through 40 µm cell strainers (Thermo Fisher). Cells that passed through the filters were incubated with anti-CD31 (PECAM1)-coated magnetic micro-beads (Miltenyi Biotech) to deplete endothelial cells (1 h at 4°C with gentle rocking). After removing magnetic beads using a Dynal magnetic bar, cells were collected from the suspension by centrifugation at 1000 g for 5 min, and re-suspended in DMEM containing 10% FBS, 10 ng/ml PDGF-A (R&D Systems, now part of Thermo Fisher), 20 ng/ml γ-interferon (ProSpec), 2 mM L-glutamine, 2 mM sodium pyruvate, 1% non-essential amino acids, 100 mg/ml streptomycin and 100 U/ml penicillin (all from Life Technologies). Re-suspended cells were plated out on feeder cells (mitomycin-treated mouse embryonic fibroblasts). Cells were incubated in a hypoxia cell culture incubator (5% O2, 5% CO2, 37°C). Astrocytes remained viable for several weeks whereas other cell types gradually died out. The astrocyte identity of the surviving cells was validated by anti-GFAP and anti-Pax2 double immunofluorescence staining. The absence of ECs was confirmed by lack of CD31+ cells.
Quantitative RT-PCR and ELISA
Total RNA was purified from cultured retinal astrocytes using TriZol (Invitrogen, now Life Technologies) according to the manufacturer's instructions. Quantitative real-time RT-PCR (qPCR) analyses were performed as described previously (Duan et al., 2011), using primers listed in Table S2. β-Actin was used as the base for normalization.
For ELISA, conditioned media were collected from retinal astrocyte cultures, and used for VEGF-A and EPO ELISA kits, respectively (R&D Systems, now part of Thermo Fisher; catalog numbers MMV00 for VEGF kit, and MEP00B for EPO kit).
Nuclear protein extracts were prepared from primary retinal astrocyte cultures for western blotting. Cells were homogenized in ice-cold Buffer A, consisting of 10 mM HEPES-KOH (pH 7.9), 10 mM KCl, 1.5 mM MgCl2, 0.5 mM dithiothreitol, 0.1% NP-40, 1× complete protease inhibitor cocktail (Roche), 0.2 mM PMSF (phenylmethylsulfonyl fluoride) and 0.2 mM deferoxamine (Sigma-Aldrich). Nuclei were pelleted by centrifugation at 4500 g for 5 min, and re-suspended in ice-cold Buffer C, which contains 20 mM HEPES-KOH (pH 7.9), 1.5 mM MgCl2, 420 mM NaCl, 0.5 mM dithiothreitol, 0.2 mM deferoxamine, 1× protease inhibitor cocktail, 0.2 mM PMSF and 25% glycerol. Antibodies are described in Table S3.
Whole-mount staining of neonatal retinas and laser confocal microscopy
IF staining of neonatal retinas was performed using antibodies validated in previous publications (Duan et al., 2017, 2014). Neonatal mice (P8 or younger) were euthanized by decapitation, and whole eyes were isolated by enucleation. Eyes were fixed immediately in 4% paraformaldehyde (PFA) for 45 min at room temperature, and then retinas were isolated. Each retina received four radial incisions, leaving four petals attached to one another at the optic nerve head. To visualize blood vessels, retinas were stained as whole mounts with IB4-Alexa Fluor 594 or IB4-Alexa Fluor 647 in retina staining buffer (RSB), which is PBS containing 1 mM CaCl2, 1 mM MgCl2, 1% Triton X-100 and 1% bovine serum albumin. Staining was carried out overnight at 4°C, with gentle rocking. Retinas were washed three times in RSB before being flat-mounted to slides. Whole-mount IF staining was carried out similarly to IB4 staining. Vascular smooth muscle cells (VSMCs) were stained by anti-actin α-smooth muscle-FITC under the same conditions as IB4 staining. Detailed information for antibodies and IB4 are provided in Table S3.
Stained retinas were flat-mounted in mounting media (50% glycerol in PBS, 1 mM CaCl2 and 1 mM MgCl2). Laser confocal microscopy was carried out with a Zeiss LSM 880, and images were processed with ZEN software (Zeiss). Analysis of z-stack confocal images was carried out with Imaris software (Bitplane).
Identification of proliferative and apoptotic ECs and vascular tip cells
To label proliferating cells, neonatal mice (P6) were injected with BrdU intraperitoneally at 120 mg/kg (Roche Applied Science). Mice were euthanized an hour later, and whole eyes were isolated and fixed in 4% PFA. Fixed eyes were stored in 70% ethanol overnight. Retinas were isolated as described in the preceding section. Isolated retinas were incubated sequentially in PBS containing 1% Triton X-100 for 30 min at room temperature, then in 2 mM HCl for 1 h at 37°C, and finally in RSB for 2 h. BrdU incorporation into vascular structures was detected by whole-mount double staining with biotin-conjugated rat anti-BrdU and Streptavidin-DyLight-549 (Table S3). Stained retinas were flat-mounted for confocal microscopy.
Apoptotic ECs were identified as IB4+/Cas3+ cells in retinas double stained with IB4-Alexa Fluor-594 and anti-caspase 3 directed to the active fragment (Cell Signaling). Secondary antibody for anti-caspase 3 was goat anti-rabbit IgG-Alexa Fluor-488. As a positive control, neonatal mice (P7) were exposed to 75% oxygen for 16 h to cause extensive EC apoptosis. Oxygen-exposed neonates were euthanized, and eyes and retinas were dissected and fixed as described in earlier sections.
Tip cells were identified essentially according to Gerhardt et al. (2003) Briefly, whole-mount retinas (P5) were double stained with SYTOX green and IB4-Alexa Fluor-594, and 6 µm z-stack images were taken. Tip cells were identified as IB4+ positive cells at the leading edge of the vasculature containing SYTOX positive nuclei, with filopodia extending further into avascular areas.
Intravitreal injection of PDGF-A and anti-VEGF
Intravitreal injection was performed with a custom tip 33 gauge needle (Hamilton) attached to a 10 µl gastight syringe (Hamilton). PDGF-A injection was performed at P4 in anesthetized mice. Each mouse was injected with 1 µl of PDGF-A (100 ng/µl in PBS, R&D Systems-Thermo Fisher) in one eye, and 1 µl of PBS in the contralateral eye. Injected mice were euthanized at P8, and retinas were dissected, fixed and stained as described in preceding sections.
Anti-VEGF was injected at 1 µl per eye (1.8 µg/µl, Cedarlane), with the contralateral eye injected with PBS as control.
Quantification of vascular content
Vascular content was quantified as number of vascular branches per area (0.02 mm2) or percentage of an area occupied by IB4+ pixels (% IB4+ area), as indicated where appropriate. Because vascular density may vary substantially at different retinal locations, all quantifications refer to specified retinal positions. To minimize sampling errors, at least two samples were quantified per specified location in each mouse, and the average of the two or more equivalent areas in each mouse was used as one data point.
Where it was impractical to directly count vascular branches because of overcrowding, the number of non-vascular spaces surrounded by blood vessels was enumerated. Mathematically, the two sets of the numbers are identical as long as all vascular branches meet at their ends to form enclosed loops, which is mostly true in the retina. At the perimeters of each specified rectangular area, every two non-vascular spaces crossing the line were counted as one. All quantifications were assisted by NIH ImageJ software.
Quantification of astrocyte development
Astrocytes were quantified in specified areas as % GFAP+ area (percentage of an area occupied by GFAP+ pixels), % Pax2+ area, and the number of Pax2+ nuclei per 0.02 mm2 area. Astrocyte differentiation was quantified as the ratio between % GFAP+ area and % Pax2+ area. Astrocyte quantification was assisted by NIH ImageJ software.
Data were evaluated by two-tailed Student's t-tests using Excel, and are presented as mean±s.e.m. All n values refer to number of mice. P<0.05 was considered significant.
We thank the Imaging Core and Molecular Core at the UConn Heath Center for technical support.
Conceptualization: G.-H.F.; Methodology: L.-J.D., G.-H.F.; Validation: L.-J.D.; Formal analysis: G.-H.F.; Investigation: L.-J.D.; Data curation: L.-J.D., G.-H.F.; Writing - original draft: G.-H.F.; Writing - review & editing: G.-H.F.; Supervision: G.-H.F.; Project administration: L.-J.D., G.-H.F.; Funding acquisition: G.-H.F.
This study was supported by the National Institutes of Health (NIH) (5R01EY019721). Deposited in PMC for release after 12 months.
Supplementary information available online at http://dev.biologists.org/lookup/doi/10.1242/dev.175117.supplemental
The authors declare no competing or financial interests.