Microglia refine developing retinal astrocytic and vascular networks through the complement C3/C3aR axis Free

The elevated metabolic rate of neural tissues, including the retina, requires the development of a complex vascular network for their growth, survival and function. The process of vascular development in neural tissue is orchestrated through highly regulated extrinsic and multicellular-derived intrinsic cues (Edwards et al., 2010; Fruttiger, 2007; Fruttiger et al., 1996; Gnanaguru et al., 2013; Hellström et al., 2007; Rymo et al., 2011; Stone and Dreher, 1987; Uemura et al., 2006a). However, the mechanism(s) that dictate the spatial assembly of vascular network formation during neural tissue development is not fully resolved. Here, we have explored the cellular and molecular contributions of the resident CNS immune cell, the microglia, in the formation of intricate and spatially organized astrocytic and vascular networks in the neuroretina during development.

Vascular development in the retina is completely dependent on astrocytes that enter the retina through the optic nerve head around birth in mice (Fruttiger et al., 1996; Stone and Dreher, 1987; Watanabe and Raff, 1988). As astrocytes proliferate and migrate into the retina, they form spatially organized honeycomb-shaped matrices that guide endothelial tip cells (starting at P0) to define the primary superficial vascular plexus (Dorrell et al., 2002; Fruttiger et al., 1996; Gerhardt et al., 2003; Uemura et al., 2006b). Upon completion of the primary superficial vascular plexus (around P7), the deep and intermediate vascular plexuses arise from the existing superficial vascular network to form an interconnected retinal vascular system (by P15) (Fruttiger, 2007). Disruption of astrocyte template assembly perturbs subsequent growth of the retinal vascular plexus (Fruttiger et al., 1996; Gnanaguru et al., 2013; Uemura et al., 2006b). Although the importance of astrocyte template formation for retinal vascular growth is known, it remains unclear how astrocytes assemble into a spatially organized honeycomb structure to guide retinal vascular development. Understanding these developmental processes is vital because in cases of perturbed or abnormal vascular growth, permanent vision loss can occur (Hartnett, 2020; Holmstrom et al., 2007; Wong et al., 2016).

During neural development, highly dynamic microglial population play an indispensable role in the formation of a functional neural circuitry mediated in part through the complement system (Anderson et al., 2019; Paolicelli et al., 2011; Schafer et al., 2012). The complement system is an integral component of the innate immune system, initially discovered as a component involved in opsonization and clearance of foreign pathogens (Holers, 2014). Recent studies imply that, in addition to its immune surveillance function, the complement system regulates a wide variety of biological activities, such as phagocytic clearance of neurons or dying cells, synaptic pruning, regulation of glial-microglial crosstalk, clearance of abnormal neovascular growth and modulation of cellular metabolism (Anderson et al., 2019; Baudino et al., 2014; Coulthard et al., 2018; Inafuku et al., 2018; Kim et al., 2016; Lian et al., 2016; Mevorach et al., 1998; Mukai et al., 2018; Roy et al., 2013; Schafer et al., 2012; Sweigard et al., 2015). Depending on the context/stimuli, complement activation is initiated through the classical, alternative or lectin pathways (Pitulescu et al., 2017), all of which lead to cleavage of C3 fragments, i.e. the large C3b and the small C3a fragments (Holers, 2014; Merle et al., 2015). In particular, during neural development, the release of cleaved C3a binding and activation of its cognate C3aR regulates diverse biological functions, such as the development of the cerebellum, modulation of embryonic neural progenitor proliferation and eye morphogenesis (Bénard et al., 2008; Coulthard et al., 2018; Grajales-Esquivel et al., 2017), suggesting that C3aR plays an important role during neurogenesis. On the other hand, the release of C3b binding and activation of CR3 (also known as integrin alpha M) facilitates microglial developmental refinement of neural tissue architecture (Anderson et al., 2019; Schafer et al., 2012). This suggests a more ubiquitous role of C3 during retinal vascular development.

In the course of retinal vascular development, a significant number of astrocytes are eliminated through an unknown mechanism (Chan-Ling et al., 2009; Puñal et al., 2019), which is vital for regulated vascular growth. Here, we report that microglia use the C3/C3aR axis (which is distinct from the C3/CR3 axis involved in phagocytic removal of neurons during development) to regulate the spatially organized astrocyte template and vascular network formation during neuroretinal development.

Astrocytes that express platelet-derived growth factor receptor α (Pdgfra) enter the retina through the optic nerve head in response to Pdgfa secreted by ganglion cells around birth in mice (Fruttiger et al., 1996, 2000). As the astrocytes migrate into the retina, they establish a honeycomb-shaped template required for endothelial tip cells to follow and form the primary superficial retinal vasculature (Fruttiger, 2007; Gerhardt et al., 2003). Defective formation of the astrocyte template severely perturbs vascular growth in the retina (Gnanaguru et al., 2013; Uemura et al., 2006b). The mechanism(s) involved in the spatial structuring of the astrocyte template required for retinal vascular development are not clearly known. To determine the role of microglia in the establishment of the spatially organized astrocyte template, we used a microglial-specific P2ry12 marker (Butovsky et al., 2014; Okunuki et al., 2018) to label microglia and characterize microglial-astrocyte interactions during the astrocyte template formation (P0) and rearrangement (P5) phases. At P0, P2ry12 labeling revealed an intricate association of microglial processes with astrocytes around the vascularizing optic nerve head (ONH) region, as well as in the avascular region containing the naïve astrocytic bed (Fig. S1A,C). Further analysis at P5 showed continuous close interactions between microglia and astrocytes in the vascular and avascular regions (Fig. S1B,C). Moreover, co-labeling of microglia with P2ry12 and Iba1 illustrated that P2ry12-labeled retinal microglia exhibited more prominent cell surface structures compared with staining with Iba1 or isolectin B4 (Fig. S1D,E), allowing us to define a clearer interaction between microglia and astrocytes during retinal vascular development than previously known.

To investigate the functional relevance of microglial-astrocyte interactions during the crucial steps of astrocyte template assembly, we designed a strategy to deplete microglia around birth. Colony-stimulating factor 1 receptor (Csf1r) signaling is required for microglial survival (Elmore et al., 2014; Okunuki et al., 2018, 2019), and previous work has found that Csf1r deletion early during development is embryonic lethal (Chitu and Stanley, 2017). Therefore, we used a pharmacological approach whereby Csf1r is inhibited around birth to deplete microglia. We have previously reported that this pharmacological approach effectively depletes microglia in the retina, similar to genetic approaches with minimal off-target effects (Okunuki et al., 2019, 2018).

P2ry12 immunostaining of P1 and P5 retinal flatmounts from the control diet-fed or the PLX5622 diet-fed group indicated that the PLX5622 antagonist successfully depleted microglia in comparison with controls (Fig. 1A,C). Analysis of RNA extracted from P1 and P5 retinas of the PLX5622 group further confirmed a significant decrease in Csf1r and microglial marker (Tmem119 and P2ry12) gene expression levels in comparison with the control group (Fig. 1B,D). After successful microglial depletion during early retinal vascular development, we examined P1 (active astrocyte template assembly phase) and P5 (active astrocyte template rearrangement and vascular growth phase) retinas of control and microglia-depleted pups.

Microglial depletion did not prevent, but moderately reduced, astrocyte migration at P1 (Fig. 1E). Although there was no significant difference in the retinal astrocyte coverage between control or PLX5622 groups at P5, retinal flatmounts lacking microglia displayed increased Sox9⁺ (astrocyte nuclear marker) (Sun et al., 2017) and Pdgfra⁺ astrocyte density in the central vascular and peripheral avascular zones in comparison with control retinas (Fig. 2A,B; Fig. S2B). Quantitative analysis by flow cytometry further revealed that the relative percentage of Pdgfra⁺ astrocytes is significantly increased in pups lacking microglia compared with the control group (Fig. 2C). In addition, microglia-depleted retinas displayed aggregation of Pdgfra⁺ astrocyte network in some areas of the central and peripheral regions compared with the control retinas at P5 (Fig. S2B). We also verified that PLX5622 treatment did not alter the expression level of the retinal astrocyte growth factor Pdgfa (Fig. 2D).

Intriguingly, the increased astrocytic density persisted even after the completion of vascular development (at P15) (Fig. S3A,B). Astrocyte distribution in P15 retinal flatmounts with and without microglia was assessed using a Gfap marker, as it is strongly expressed in mature astrocytes (Chan-Ling et al., 2009). Gfap-immunostained P15 retinas of pups lacking microglia showed a significant increase in astrocyte distribution in the central and peripheral retinas compared with controls (Fig. S3A,B).

Microglial depletion resulted in defective assembly of the astrocytic template, so we next examined the effects on subsequent vascular growth. Analysis of P1 and P5 retinal flatmounts in litters from mice lacking microglia showed a significant decrease in vascularized area in comparison with the control group (Fig. 3A,B). Further analysis of retinal flatmounts during (P5) and after (P15) the completion of vascular development displayed a significant reduction in vascular branch points and density in comparison with the controls (Fig. 3C,D, and Fig. S3C-E). We also examined whether the PLX5622 inhibitor had any off-target effect on suppressing Vegf isoform expression levels. Data show that PLX5622 treatment did not suppress Vegfa or Vegfc isoform levels (Fig. 3E). As vascular growth and branching are reduced in the microglia depleted retinas, the moderate increase of Vegfa expression observed in PLX5622 could be a compensatory mechanism. Taken together, these results implicate microglia in the proper development of the astrocytic and vascular networks (Fantin et al., 2010; Outtz et al., 2011; Rymo et al., 2011).

Given that microglial depletion significantly increased astrocyte density and altered spatial patterning (Fig. 2), we next focused on characterizing the mechanism through which microglia regulate astrocyte template patterning. During retinal vascular development, large numbers of astrocytes are eliminated at P5 through an uncharacterized mechanism facilitated by microglia (Chan-Ling et al., 2009; Puñal et al., 2019). In agreement with previous findings (Puñal et al., 2019), analysis of P5 retinal flatmounts exhibited microglia-engulfed Gfap-positive astrocyte cellular debris localized within the microglial endosomal/lysosomal membrane protein CD68 (Fig. S4A).

During retinal development, microglia facilitate the elimination of dying neurons and synapses through complement components C1q and C3 (Anderson et al., 2019; Schafer et al., 2012; Stevens et al., 2007). We therefore examined whether astrocyte and vascular densities are altered in C1q knockout mice and found that the loss of C1q did not alter astrocyte and vascular patterning at P5 (Fig. S4B,C).

Next, to investigate whether microglial removal of astrocytes during retinal development is C3 dependent, we examined C3 expression during the active astrocyte elimination phase (P5) using C3 tdTomato reporter mice. Examination of retinal cross-sections revealed C3 tdTomato expression in microglia at the astrocyte layer and in the underlying ganglion cells (Fig. 4A). Further analysis of P5 retinal flatmounts of C3 tdTomato reporter mice showed robust C3 tdTomato localization around the vascular growth and adjacent avascular zones (Fig. 4B). High resolution images of retinal flatmounts of P5 C3 tdTomato reporter mice revealed that C3 tdTomato was predominantly localized in microglia that are located at the astrocyte layer (Fig. 4C). Quantification of C3 tdTomato expression in microglia revealed that about 15% of the cells expressed C3 tdTomato (Fig. 4D).

In order to determine whether C3 facilitates restructuring of the astrocyte template during vascular development, we analyzed P5 retinal flatmounts of wild-type and C3-deficient littermates. Analysis of C3-deficient retinas showed increased astrocyte nuclear marker Sox9 distribution and increased Pdgfra⁺ astrocyte network density in the vascular and avascular zones (Fig. 4E,F). Furthermore, we did not see any significant increase in Pdgfa transcript levels (Table S2) or changes in axonal density (Fig. S5A), which are reported to regulate astrocyte proliferation and distribution (Fruttiger et al., 2000; O'Sullivan et al., 2017). We then analyzed whether microglial phagocytic removal of astrocytes is altered in C3-deficient retinas. Data showed a significant reduction in the localization of Gfap-positive astrocyte debris within microglial CD68 in C3-deficient retinas compared with wild-type littermate control retinas (Fig. 5A and Fig. S5B), indicative of defective microglial phagocytic clearance of astrocytes. These results suggested a stronger selective role for C3 in regulating astrocyte template patterning during retinal vascular development.

We next analyzed the effects of C3 loss on subsequent vascular growth at P5. In comparison with wild-type littermates, P5 C3-deficient retinal flatmounts showed a significant increase in the vascularized area and density (Fig. 5B,C). In order to gain further mechanistic insight that could explain the observed increased astrocyte density and enhanced vascular growth phenotypes in C3-deficient retinas, we performed bulk mRNA sequencing. RNA-seq results show that 451 genes were significantly differentially expressed in P5 C3-deficient retinas in comparison with P5 wild-type retinas (Table S2). To identify the pathway(s) altered in C3-deficient retinas during vascular development, we next performed pathway enrichment analysis on all the significantly differentially expressed genes using EnrichR (Chen et al., 2013). Data analysis shows that the top ten pathways elevated in C3-deficient retinas included enrichment of genes linked to extracellular matrix and integrin signaling necessary for cellular interactions and migration during angiogenesis (Fig. 5D,E). Notably, ECM-associated genes that were elevated in C3-deficient retinas [Lamc3, Col4a3, Col4a4, Fmod, Fbn1, Bmp4, Col8a1, Col8a2 and Col1a2 (Fig. 5D)] are all known to promote angiogenesis and vascular stability (Adini et al., 2014; Bao et al., 2021; Chen et al., 2022; Gnanaguru et al., 2013; Jian et al., 2013; Lopes et al., 2013; Rajan et al., 2020; Rezzola et al., 2019). It is also noteworthy that Pax2 (an astrocyte marker) was among the genes elevated in C3-deficient retinas (Fig. 5D). These results suggest that C3 plays an important regulatory role in facilitating astrocyte template reorganization that is necessary for regulated subsequent superficial retinal vascular growth.

Microglia express the C3 receptors CR3 and C3aR that bind cleaved C3b cleaved C3a peptide, respectively (Lian et al., 2016; Schafer et al., 2012). To determine which cleaved C3-binding receptor(s) is involved in the regulation of astrocyte and vascular patterning, we examined the gene expression levels of CR3 and C3aR in the microglia-depleted retinas. Gene expression analysis showed that both CR3 and C3aR are significantly downregulated in microglia-depleted retinas compared with the control retinas (Fig. S5C). During retinal development, cleaved C3b fragment binding and activation of CR3 is shown to facilitate microglial removal of ganglion cells (Anderson et al., 2019). We therefore examined whether C3 acts through CR3 in regulating astrocyte and vascular patterning. Our analysis of P5 CR3 KO retinal flatmounts did not show significant changes in astrocyte patterning or vascular growth (Fig. S5D,E), suggesting that other C3 cleavage fragments and their cognate receptors are involved in retinal vascular development. We next investigated whether there is a possible role for C3a/C3aR in facilitating microglial phagocytic elimination of astrocytes during retinal vascular development. Analysis of P5 retinal flatmounts also showed deposition of cleaved C3a over the surface of the astrocyte network (Fig. S6A). Further examination of P5 retinal flatmounts of C3aR-tdTomato reporter mice (Quell et al., 2017) showed that all microglia expressed C3aR tdTomato (Fig. 6A-C), with the strongest expression in the vascular growth and adjacent avascular zones (Fig. 6A). Moreover, examination of C3aR KO retinas showed a significant increase in the astrocyte nuclear maker (Sox9) and network density in the vascular and avascular zones compared with wild-type retinas (Fig. 6D,E). Assessment of Pdgfa transcript level and axonal density did not show any significant change in wild-type and C3aR KO retinas at P5 (Fig. S6B,C). Further examination of P5 C3aR KO retinal flatmounts showed a significant reduction in the localization of Gfap-positive astrocytic cell debris within microglial CD68 compared with wild-type littermate control retinas (Fig. 7A and Fig. S6D). The anaphylatoxin receptor C3aR is not largely known for its phagocytic function. To determine whether C3aR carries phagocytic function, we isolated and cultured retinal microglia and astrocytes from P5 pups. After labeling cultured astrocytes with DiI for tracking purposes, cell death was induced by staurosporine treatment. Retinal microglia were incubated with DiI-labeled dead astrocytic bodies in the presence or absence of C3aR neutralizing antibodies or control IgG. Blocking C3aR significantly reduced microglial engulfment of DiI-labeled dead astrocytic bodies when compared with the IgG control (Fig. 7B), revealing the phagocytic capability of microglial C3aR.

Moreover, subsequent superficial vascular growth and vessel density were increased in P5 C3aR KO retinal flatmounts compared with wild-type littermate controls with no significant change in Vegfa mRNA levels (Fig. 7C,D and Fig. S6E). We also analyzed whether proangiogenic ECM genes expression levels were altered in P5 C3aR KO retinas, similar to C3 KO (Fig. 5D). A gene expression study showed that the expression levels of Pax2, Col4a3, Col8a1 and Fmod were significantly increased in C3aR KO retinas, similar to C3-deficient retinas (Figs 7E and 5D). There was also an upward trend in Lamc3, Col4a4, Col8a2 and Fbn1 gene expression levels in P5 C3aR KO retinas compared with wild type (Fig. 7E). Of note, we were unable to perform flow cytometry on P5 C3- and C3aR-deficient retinas, as enzymatic digestion often resulted in clustering of retinal cells in suspension, likely due to increased accumulation of ECM proteins (Figs 5D and 7E).

Analysis of P15 (vascular development completion stage) C3aR-deficient retinas displayed areas of abnormal Gfap⁺ astrocyte accumulation in several areas (Fig. S7A). Although the vascular density is not significantly increased (Fig. S7B), C3aR-deficient retinas displayed the presence of dysmorphic non-vasculature associated CD31⁺ endothelial cells in a number of regions at the superficial vascular layer at P15 (Fig. S7B). These results demonstrate that microglial C3aR is necessary for spatial restructuring of the astrocyte template and the superficial vascular network. In summary, our study identifies that microglia, through the C3a/C3aR axis, regulate the astrocyte spatial patterning required for subsequent vascular development.

Astrocyte migration and template formation is an essential step for vascular development in the retina. Interestingly, species that lack astrocytes in the retina remain avascular (Schnitzer, 1988; Stone and Dreher, 1987; Zhang and Stone, 1997). The entry and migration of astrocytes into the retina is dependent upon ganglion cell and ECM-derived cues (Fruttiger et al., 1996, 2000; Gnanaguru et al., 2013; Gnanaguru and Brunken, 2012; Tao and Zhang, 2016). While migrating towards the periphery of the retina, astrocytes form spatially organized ECM enriched matrices that sequester growth factors for the purpose of guiding endothelial cells (Dorrell et al., 2002; Gerhardt et al., 2003; Uemura et al., 2006b). Until now, it had been unclear how astrocytes assemble into spatially organized structures to guide retinal vascular growth. In this study, we demonstrate that microglia closely associate with astrocytes during template formation. Depletion of microglia before the entry of astrocytes into the retina severely altered astrocyte template formation and increased cellular density at P5. Supporting this finding, depletion of microglia around P4-P6 using a genetic approach also increased astrocyte density (Puñal et al., 2019). This suggests that microglia play a crucial role in spatial structuring of astrocyte template.

Additional studies have uncovered a mechanism through which microglia facilitate astrocyte template organization during vascular development in the neuroretina. Results from our study show that a unique complement axis facilitates microglial phagocytic removal of astrocytes during retinal vascular development for the spatial establishment of organized astrocyte and vascular networks. Previous studies indicate that microglia through the C3b/CR3 and C1q complement axis regulate developmental pruning and restructuring of retinal neurons/synapses (Anderson et al., 2019; Cong et al., 2020; Schafer et al., 2012). Our analysis of CR3- or C1q-deficient retinas did not show any significant changes in astrocyte spatial patterning. However, assessment of C3-deficient retinas showed a significant increase in astrocyte spatial density and defective phagocytic removal of astrocyte bodies during retinal vascular development. These findings suggest that perhaps C1q and C3b/CR3 are more specific for phagocytic removal of neurons and pruning of synapses (Anderson et al., 2019; Schafer et al., 2012), while the microglial phagocytic elimination of astrocytes during retinal vascular development involves the distinct C3 axis (i.e. C3a/C3aR).

The C3a/C3aR axis is widely regarded as a complement activation system involved in the immune response (Harder et al., 2020; Ingersoll et al., 2010; Lian et al., 2016). Apart from immune surveillance function, the C3a/C3aR axis plays a role in regulating neural development (Coulthard et al., 2018). Our studies unveiled a new regulatory function of the C3/C3aR axis during vascular development in the neuroretina. Analysis of C3- and C3aR-deficient retinas displayed diminished microglial phagocytic removal of astrocytes, leading to an increase in astrocyte density. Moreover, C3aR-deficient retinas showed abnormal accumulation of astrocytes in some areas, even after the completion of retinal vascular development (P15). We did not detect such abnormal accumulation of astrocytes in C3-deficient retinas at P15 (data not shown). It remains to be elucidated whether other complement pathways (Dong et al., 2017; Nauser et al., 2018) compensate for C3 loss by facilitating microglial elimination of astrocytes during the completion of retinal vascular development. Nevertheless, C3aR-deficient retinas continued to display abnormal accumulation of astrocytes, suggesting an important regulatory role played by the C3aR axis in facilitating astrocyte spatial organization during retinal vascular development.

While migrating into the retina, astrocytes assemble into honeycomb-shaped ECM-enriched matrices that sequester growth factors, such as vascular endothelial growth factor for vascular growth (Fruttiger, 2007; Gerhardt et al., 2003; Uemura et al., 2006b). Through integrin-dependent cell adhesion mechanism, the leading vascular tip cells interact and follow the astrocyte laid matrices (Park et al., 2019; Stenzel et al., 2011; Uemura et al., 2006b; Zovein et al., 2010). ECM-enriched astrocyte matrices not only serve as a reservoir for growth factor signaling, but also play an important role in directing polarized endothelial migration through integrin signaling during angiogenesis (Park et al., 2019; Ruiz de Almodovar et al., 2010; Zovein et al., 2010). Loss or accumulation of ECM proteins could, respectively, suppress or enhance endothelial migration and vascular growth (Edwards et al., 2010; Feng et al., 2013; Gnanaguru et al., 2013; Ishihara et al., 2018) by modulating cell-matrix interactions.

It is possible that the increased vascular growth observed in C3- and C3aR-deficient retinas could be due to increased astrocyte-derived ECM-enriched matrices leading to enhanced endothelial migration (Feng et al., 2013; Ishihara et al., 2018). Supporting this notion, our RNA-seq and gene expression studies show that the loss of C3 and C3aR led to increased proangiogenic ECM protein gene expression levels. Increased deposition of proangiogenic matrix proteins, such as isoforms of laminin, collagen and fibromodulin, would ultimately lead to more pronounced endothelial migration and vascular growth, in part through integrin-dependent signaling (Chen et al., 2022; Gnanaguru et al., 2013; Ishihara et al., 2018; Jian et al., 2013; Marchand et al., 2019; Park et al., 2019; Zovein et al., 2010).

In contrast to C3- and C3aR-deficient retinas, vascular network formation was significantly reduced in microglia-depleted retinas, even though the astrocyte density was increased. During endothelial tip cell migration over the surface of astrocytes, microglia help the fusion of vascular tip cells through anastomoses to create a complex vascular network (Fantin et al., 2010; Haupt et al., 2019; Outtz et al., 2011; Rymo et al., 2011). Our results clearly indicate that ,without microglia, a well-defined complex vascular network cannot be formed, even if the astrocyte density is increased. The deficiency of C3 and C3aR did not result in microglial loss or disrupted microglial-astrocyte-endothelial interactions during retinal vascular development (only phagocytic elimination of astrocytes by microglia was reduced). In fact, our RNA-seq study illustrated an increase of the microglial marker Fcrls in C3 deficient retinas (Fig. 4E). Therefore, the increased vascular network seen in C3- and C3aR-deficient retinas could possibly be due to increased microglial fusion of vascular tip cells that are migrating on densely distributed matrix-enriched astrocytes.

Taken together, our findings provide compelling evidence for a novel functional role for the microglia-derived C3/C3aR axis in creating spatially organized astrocyte matrices to fine-tune superficial retinal vascular morphogenesis during retinal development.

All animal procedures were performed in accordance with the Massachusetts Eye and Ear Animal Care Committee. Mice were maintained in a room with a 12 h light/12 h dark cycle. C57BL6/J (stock 00664), C3 mutant (referred to as KO) (stock 029661), C3aR mutant (referred to as KO) (stock 033904), C1q KO (stock 031675), CR3 KO (also known as Itgam) (stock 003991) breeding pairs were purchased from Jackson Laboratories. C1q KO, CR3 KO, C3 KO and C3aR KO mice were crossed to C57BL6/J mice to create a heterozygous line. Pups generated from the heterozygous lines carrying C1q^+/+ (wild type), C1q^−/−, CR3^+/+, CR3^−/−, C3^+/+, C3^−/−, C3aR^+/+ and C3aR^−/− genotypes were used for the studies. Of note, it is reported that mutant C3 mRNA was detected in extra-hepatic tissues of deficient mice with no detectable level of active cleaved C3 proteins (Circolo et al., 1999; Wessels et al., 1995), and our RNA-seq analysis also showed elevated mutant C3 mRNA expression in C3 KO (Table S2). Floxed C3 IRES-tdTomato reporter mice (Kolev et al., 2020), obtained from Dr Kemper (Immunology Center, National Heart, Lung, and Blood Institute, NIH, Bethesda, USA), were bred to C57BL6/J mice and maintained on a C57BL6/J background. Floxed C3aR-tdTomato reporter knock-in mice (Quell et al., 2017), obtained from Dr Köhl (Institute for Systemic Inflammation Research, University of Lübeck, Lübeck, Germany), were maintained on a C57BL6/J background.

Timed-pregnant mice were maintained on a control chow diet or a chow diet containing Csf1r inhibitor (formulated at 1200 ppm) (PLX5622, Plexxikon) from gestational day 13.5 or 14.5. Retinas were collected from pups at postnatal days (P) 1, P5 and P15 for downstream analysis.

Astrocytes and microglia were isolated using immunopanning methodology (Collins and Bohlen, 2018; Zhang et al., 2016). In brief, six-well culture plates were coated with their respective secondary antibodies (anti-rat IgG, 312-005-045, Jackson ImmunoResearch or sheep IgG, 5-001-A, R&D Systems) [10 µg/ml diluted in sterile 50 mM Tris-HCl (pH 9.5)] by incubating for 1 h at 37°C. After washes, culture plates were coated with 2.5 µg/ml of anti-CD11b (MAB1124, R&D Systems) or anti-integrin β5 (AF8035, R&D Systems) antibodies overnight in the cell culture hood and washed in 1×PBS. P5 retinas (n=8) were isolated in ice-cold sterile 1×DPBS (Thermo Fisher Scientific) and incubated in DMEM/F12 containing 5% fetal bovine serum (FBS), collagenase D (1 mg/ml) and DNase (0.1 mg/ml) for 30 min at 37°C. The retinas were then gently triturated in DMEM/F12 supplemented with 10% FBS and filtered through a 40 µm cell strainer to remove cell clumps and debris. The resultant single cell suspension was incubated in a secondary antibody-coated well for 10 min at 37°C to deplete non-specific cell binding. The unbound cell suspension was then transferred to anti-CD11b-coated wells and incubated for 20 min at 37°C to capture microglia (the culture plate was shaken every 5 min to dislodge non-specific cell binding). After depleting microglia, the cell suspension was transferred to integrin β5-coated wells to capture astrocytes and incubated for 45 min at 37°C (culture plates were shaken every 10 min to dislodge non-specific cell binding). After removing the unbound cells, microglial and astrocyte cell culture plates were thoroughly washed in serum-free DMEM/F12. Microglial cells were grown and maintained in DMEM/F12 supplemented with 5% FBS and 10 ng/ml of recombinant mouse Csf1, and astrocytes were grown and maintained in DMEM/F12 supplemented with 10% FBS and N-2.

To examine microglial phagocytosis of dead astrocyte bodies, ∼2500 microglial cells were plated on a 12 mm glass coverslip and maintained for 48 h. Astrocyte cell suspensions were labeled with Vybrant Dil cell labeling solution as suggested by the manufacturer (Thermo Fisher Scientific) and cell death was induced by incubating the labeled astrocyte cell suspension with 1 µM staurosporine (Cayman Chemical Company). Microglial cells were then treated with DiI-labeled dead astrocyte bodies in a 1:1 ratio and incubated for 2 h. For receptor neutralization assays, microglial cells were pre-incubated with 20 µg/ml of anti-C3aR antibody (MAB10417) or control IgG (R&D Systems) suspended in DMEM/F12 supplemented with 5% heat-inactivated FBS, followed by treatment with DiI-labeled astrocyte dead bodies in the presence of antibodies for 2 h at 37°C. After thorough washes, cells were fixed in 4% PFA, blocked for 1 h at room temperature (in 10% fetal bovine serum, 0.05%triton-X100 and 0.01% sodium azide in 1×PBS), followed by incubation with P2ry12 antibody (AS-55043A, AnaSpec) overnight at 4°C and then incubated with goat anti-rabbit Alexa Fluor 488 or 647 (A21206 or A21447, 1:500, Thermo Fisher Scientific), images were acquired using an epifluroscent microscope (Axio Observer Zeiss).

Retinal flatmounts were prepared as described previously (Gnanaguru et al., 2013). In brief, eyes were fixed in 4% paraformaldehyde (PFA) for 10 min and then dissected in 1×PBS. After discarding the anterior chamber and lens, the retina was separated from the sclera. Retinal flatmounts were prepared with four radial incisions and the flatmounts were then stored in methanol at −20°C until immunostaining was performed. For immunohistochemistry, retinas were rehydrated in 1×PBS, washed three times and blocked in blocking buffer (10% fetal bovine serum, 0.05% Triton X-100 and 0.01% sodium azide in 1×PBS) for 2 h at room temperature. After blocking, retinas were then incubated with primary antibodies for 24 h at 4°C. After washes in 1×PBS, retinal flatmounts were incubated with respective secondary antibodies for 4 h at room temperature or overnight at 4°C. The flatmounts were then washed in 1×PBS and mounted onto a slide using anti-fade medium (Permaflour, Thermo Fisher Scientific).

Primary antibodies used were as follows: rat anti-Pdgfra (1:250, CD140A, clone APA5, BD Biosciences), anti-chicken GFAP (1:1000, SKU: GFAP, Aves Labs), rabbit anti-P2ry12 (1:500, a gift from Dr Butovsky, Brigham and Women's Hospital, Boston, MA, USA), Alexa Fluor 594 anti-tubulin β3 antibody (0.5 µg/ml, 657408, BioLegend) and rat anti-mouse CD68 (1:200, clone FA-11, Biolegend), anti-rabbit Iba1 (1:500, 019-19741, FUJIFILM Wako Chemicals), chicken anti-mouse C3a (1:250, PA1-30601, Thermo Fisher Scientific) and goat anti-CD31 (1:750, AF3628, R&D Systems). Secondary antibodies used (all 1:500 dilution) were: donkey anti-chicken-594 (20167, Biotium), donkey anti-rat-594 (A21209, Thermo Fisher Scientific), donkey anti-rabbit-488 (A21206, Thermo Fisher Scientific) and donkey anti-goat-647 (A21447, Thermo Fisher Scientific).

The enucleated eyes were fixed in 4% PFA for 2 h at room temperature. The dissected posterior eye-cups were cryopreserved in 10%, 20% and 30% sucrose and then frozen in Tissue-Tek O.C.T compound (Ted Pella). The eyecups were then cut at 12 µm and used for immunohistochemistry.

Singe cell suspensions were prepared from retinas that were dissected in ice-cold HBSS and then incubated with collagenase D (1 mg/ml) and DNAse (0.1 mg/ml) (Millipore Sigma) in DMEM/F12 supplemented with 5% FBS for 30 min at 37°C. After filtering the cells through a 40 µm cell strainer, single cell suspensions were blocked with anti-mouse CD16/32 monoclonal antibody (0.5 µg/ml, 14-0161-82, Thermo Fisher Scientific) and then stained with APC anti-Pdgfra (CD140a, clone APA5, 1:50) and PE anti-p2ry12 (clone S16007D, BioLegend, 1:50). Dead cells were distinguished with DAPI stain. The data for flow cytometry were acquired using a Cytoflex S (Beckman Coulter) and analyzed using FlowJo version 10.1.

Retinas were isolated in ice-cold PBS and RNA was extracted using RNA STAT-60 (Amsbio), as recommended by the manufacturer. After determining the RNA concentration using NanoDrop, 500 ng of total RNA was used for cDNA synthesis using SuperScript IV VILO Master Mix. Real-time PCR reactions were performed in the CFX384 Real-time PCR platform (Bio-Rad) using SYBR Green master mix (Applied Biosystems) to determine the relative expression level of Pdgfa, P2ry12, Tmem119, Csf1r, Pax2, Vegfc, Vegfa, Col1a2, Fmod, Col8a1, Col8a2, Fbn1, Lamc3 and Bmp4. Ppia was used as a housekeeping gene and all primer sequences are listed in Table S1.

RNA was extracted as described above from P5 wild-type and C3-deficient retinas, and the integrity of RNA was assessed by Bioanalyzer (Agilent 2100). RIN values for all the samples used for downstream cDNA library construction were >9.0. To determine global gene expression changes, mRNA was purified with oligo d-T attached magnetic beads (NEBNext) and 500 ng of RNA was used for cDNA library construction with NEBNext ultra II directional RNA library prep kit for Illumina, as recommended by the manufacturer (New England Biolabs). cDNA libraries were validated on the TapeStation (Agilent) and the sequencing was performed with the Illumina Nextseq 2000 platform with a target of 25 million reads per sample.

Transcriptome mapping was performed using the STAR aligner (Dobin et al., 2013) and the mm9 assembly of the mouse reference genome. Read counts for individual transcripts were obtained using HTSeq (Anders et al., 2015) and the GENCODE M1 (NCBIM37) gene annotation. Differential expression analysis was performed using the EdgeR package (McCarthy et al., 2012) after normalizing read counts and including only genes with CPM>1 for at least one sample. Differentially expressed genes were defined based on the criteria of >1.5-fold change in normalized expression value and false discovery rate (FDR) <0.05. Heatmaps and PCA plots were generated using normalized gene expression values (log2 FPKM). Heatmaps were generated using the R package heatmap (Kolde, 2019). Analysis of enriched functional categories among detected genes was performed using EnrichR (Chen et al., 2013).

Samples were imaged using an epifluroscent microscope (Axio Observer Zeiss) or confocal microscopy (SP8, Leica). For 3D image reconstruction of z-stack images, the Amira 2019.4 software tool was used.

Tiled images of the entire retinal surface immunostained for CD31 were acquired. Using NIH-ImageJ (version 2.1.0/1.55c) software, scale measurements were set, and an outline of vascular growth front was created using the freehand selection tool to measure the total CD31 immunostained area.

Images of retinal flatmounts co-labeled with CD31 and P2ry12 or with CD31 and Iba1 were acquired in all the retinal quadrants near the tip cell and avascular peripheral regions by confocal microscopy. The outline of the P2ry12- or Iba1-stained cell area was created using the ImageJ (version 2.1.0/1.53c) freehand selection tool, and immunostained cell area was measured and quantified.

The astrocyte- and blood vessel-covered areas were quantified using the ImageJ (version 2.1.0/1.53c) Angiogenesis Analyzer plugin tool. In brief, images stained for Pdgfra or CD31 were opened using AngioTool, scale measurement was entered and the were processed. Automated software generated data measurements containing vascular density, branching index and lacunarity that were plotted.

Images stained for Sox9 were acquired in the vascular and avascular areas. Using ImageJ (version 2.1.0/1.53c) images were converted to 8-bit images. After background subtraction, a threshold was set, and, using the analysis function, particle size was set and measured.

Tiled images of the entire C3 tdTomato-expressing retinal flatmounts immunostained for P2ry12 were taken and, using ImageJ (version 2.1.0/1.53c), all the microglia were counted similar to Sox9 density. C3 tdTomato co-labeled P2ry12 cells were then counted using ImageJ multipoint tool to calculate the percentage of C3 tdTomato⁺ microglia in the retina at P5.

C3aR tdTomato was expressed by all microglia; therefore, images were acquired in the vascular and avascular regions of the retina and manually counted the C3aR tdTomato-P2ry12⁺ microglia using the ImageJ multipoint selection tool.

All the statistical analyses were performed using Prism version 9 software. All data are presented as mean±s.e.m. Statistical differences between the two groups were determined using an unpaired t-test. The number of samples (n) and the level of statistical significance are provided in the legend of each figure.

We thank Plexxikon for providing the PLX5622 chow diet. We sincerely thank Dr Butovsky (Department of Neurology, Brigham and Women's Hospital, Harvard Medical School) for providing the P2ry12 antibody and Dr Claudia Kemper (Immunology Center, National Heart, Lung, and Blood Institute, NIH, Bethesda, MD, USA 20892) for providing the Floxed C3 IRES-tdTomato reporter mice. We thank Nathaniel Rowthorn-Apel for assisting (G.G.) with sample collection. We thank MGH Nextgen sequencing core staff Danning Zhou and Ulandt Kim for the help with cDNA library construction and sequencing.