Mechanisms underlying microglial colonization of developing neural retina in zebrafish

Microglia are brain-resident macrophages that function as the first line of defense in brain. Embryonic microglial precursors originate in peripheral mesoderm and migrate into the brain during development. However, the mechanism by which they colonize the brain is incompletely understood. The retina is one of the first brain regions to accommodate microglia. In zebrafish, embryonic microglial precursors use intraocular hyaloid blood vessels as a pathway to migrate into the optic cup via the choroid fissure. Once retinal progenitor cells exit the cell cycle, microglial precursors associated with hyaloid blood vessels start to infiltrate the retina preferentially through neurogenic regions, suggesting that colonization of retinal tissue depends upon the neurogenic state. Along with blood vessels and retinal neurogenesis, IL34 also participates in microglial precursor colonization of the retina. Altogether, CSF receptor signaling, blood vessels, and neuronal differentiation function as cues to create an essential path for microglial migration into developing retina.


Introduction
Microglia are the resident macrophages of brain. These dedicated CNS phagocytes form the innate immune system of embryonic and adult brain. Microglia eliminate cellular debris to prevent neuroinflammation and to promote neuronal protection in vertebrates (Ashwell, 1991;Calderó et al., 2009;Lawson et al., 1990;Neumann et al., 2009;Sierra et al., 2010). They also prune unnecessary synapses to establish functional, mature neural circuits during brain development, performing a variety of cellular functions (Paolicelli et al., 2011;Tremblay et al., 2010). In contrast to other CNS cells, like neurons and astrocytes, microglia do not originate from neural plate, but are derived from mesoderm (Ashwell, 1991;Boya et al., 1979) through hematopoiesis (Ginhoux et al., 2013). In developing zebrafish, embryonic hematopoiesis occurs in successive waves that are separated anatomically and temporally. The primitive or first wave of microglial precursors is generated from myeloid cells originating in the rostral blood island (RBI) at about 11 hr post-fertilization (hpf) (Stachura and Traver, 2011;Xu et al., 2012). The definitive wave is contributed by the ventral wall of the dorsal aorta (VDA), giving rise to hematopoietic stem cells (HSCs) (Xu et al., 2015). In addition, a short intermediate wave also originates from the posterior blood island (PBI) (Bertrand et al., 2007). After 2 weeks eLife digest The immune system is comprised of many different cells which protect our bodies from infection and other illnesses. The brain contains its own population of immune cells called microglia. Unlike neurons, these cells form outside the brain during development. They then travel to the brain and colonize specific regions like the retina, the light-sensing part of the eye in vertebrates.
It is poorly understood how newly formed microglia migrate to the retina and whether their entry depends on the developmental state of nerve cells (also known as neurons) in this region. To help answer these questions, Ranawat and Masai attached fluorescent labels that can be seen under a microscope to microglia in the embryos of zebrafish. Developing zebrafish are transparent, making it easy to trace the fluorescent microglia as they travel to the retina and insert themselves among its neurons.
Ranawat and Masai found that blood vessels around the retina act as a pathway that microglia move along. Once they reach the retina, the microglia remain attached and only enter the retina at sites where brain cells are starting to mature in to adult neurons. Further experiments showed that microglia fail to infiltrate and colonize the retina when blood vessels are damaged or neuron maturation is blocked.
These findings reveal some of the key elements that guide microglia to the retina during development. However, further work is needed to establish the molecular and biochemical processes that allow microglia to attach to blood vessels and detect when cells in the retina are starting to mature.
RGCs and bipolar/amacrine cells form the inner plexiform layer (IPL). Cell fate determination is less dependent on the cell lineage of retinal progenitor cells, suggesting that both extrinsic and intrinsic mechanisms influence the status of retinal progenitor multipotency, leading to generation of diverse retinal cell types (He et al., 2012). These developmental profiles of retinal neurogenesis and cell differentiation may be coupled with microglial colonization. Although apoptosis and CSF-R signaling are suggested in microglial colonization of the retina in zebrafish (Wu et al., 2018), the mechanism underlying microglial colonization of the retina remains to be determined.
In this study, using zebrafish, we examined the developmental profile of retinal colonization by microglia precursors. The number of ocular microglial precursors progressively increases from 32 to 54 hpf. Most microglial precursors do not proliferate, suggesting that microglial colonization of the retina depends on cell migration from outside the optic cup. We found three guidance mechanisms driving microglial precursor colonization of the retina. First, IL34 initiates microglial precursor movement from yolk toward the brain and the retina. Second, microglia precursors enter the optic cup via ocular hyaloid blood vessels in the choroid fissure, suggesting that these blood vessels guide microglia to the retina. Third, microglial precursors infiltrate the neural retina preferentially through the neurogenic region, suggesting that the neurogenic state of retinal tissue acts as an entry signal for microglial precursors to infiltrate the retina. Thus, a series of guidance mechanisms promote microglial colonization from yolk to the neural retina in zebrafish.

Embryonic microglial precursors progressively colonize developing zebrafish retina
In zebrafish, early macrophages are generated from myeloid cells originating in the RBI around 11 hpf and these macrophages colonize the brain and retina by 55 hpf (Xu et al., 2015). Around 60 hpf, brain and retina-resident macrophages undergo a phenotypic transition, which indicates expression of mature microglial markers, such as apolipoprotein E (apo E) and phagocytic behavior toward dead cells (Herbomel et al., 2001). Importantly, early macrophages outside the brain never express apo E (Herbomel et al., 2001), suggesting that only brain and retina-resident macrophages give rise to microglia. Thus, early macrophages localized in the brain and retina by 60 hpf are generally accepted as microglial precursors in zebrafish. In this study, we focused on two macrophage markers, macrophage expressing gene 1.1 (mpeg1.1) (Ellett et al., 2011) and microfibrillar-associated protein 4 (mfap4) (Walton et al., 2015), and define mpeg1.1; mfap4-positive cells inside the optic cup as microglial precursors colonizing the zebrafish retina.
To ascertain how microglia precursors migrate from peripheral tissues into the neural retina during development, we generated a zebrafish transgenic line, Tg[mpeg1.1:EGFP], using the original DNA construct (Ellett et al., 2011). As previously reported (Ellett et al., 2011), our established transgenic line visualized ocular microglial precursors and enabled us to monitor their number and location in the optic cup from 24 to 54 hpf. Accordingly, we obtained 3D images using confocal laser scanning microscopy (LSM) ( Figure 1A and B). The first microglial precursor cells appeared near the choroid fissure and lens around 30-32 hpf. After that, the number of ocular microglial precursors increased to 19.1 ± 1.26 at 42 hpf and 31.0 ± 4.44 at 54 hpf ( Figure 1C), indicating a progressive increase in the number of ocular microglial precursors. We also confirmed a similar progressive increase in the number of ocular L-plastin-positive cells, although L-plastin is expressed in microglial precursors and neutrophils in zebrafish brain (Figure 1-figure supplement 1). Next, to determine more precisely the spatial distribution of microglial precursors in the optic cup, we generated another transgenic line, Tg[mfap4:tdTomato-CAAX], using the original DNA construct (Walton et al., 2015). As previously reported (Walton et al., 2015), our established transgenic line efficiently labeled ocular microglial precursor membranes. We labeled this transgenic embryo using Bodipy ceramide conjugated with fluorescent Alexa-488, which visualizes retinal layer structures (Figure 1-figure supplement 2). From 32 to 36 hpf, mfap4 + cells were mostly located in the vitreous space between the neural retina and lens, and possibly associated with ocular blood vessels, which develop around the lens. In 42-44-hpf retina, a few microglial precursor cells start to enter the neural retina and spread toward the emerging IPL, where they are associated with newly born amacrine cells (Figure 1-figure supplement 3). By 54 hpf, IPL formation is complete and microglial precursors were observed throughout all retinal tissue,  except the OPL. Thus, microglial precursors enter the optic cup along the choroid fissure at 30 hpf, remain temporarily in the vitreous space between the lens and the retina, and then begin spreading into differentiating retinal tissue after 42 hpf.
Next, to evaluate the contribution of cell proliferation to the increasing number of ocular microglial precursors, we labeled ocular microglial precursors with markers of DNA replication. Here, we used a zebrafish transgenic line, Tg[EF1α: mCherry-zGem] that specifically marks proliferative cells in S and G2 phases (Mochizuki et al., 2017;Mochizuki et al., 2014). We combined this Tg[EF1α: mCherry-zGem] system with Tg[mpeg1.1:EGFP] to calculate the fraction of proliferative microglial precursors undergoing S phase ( Figure 1D and Video 1). First, we observed mCherry-zGem; mpeg1.1:EGFP double-positive cells in the peripheral tissue ( Figure 1-figure supplement 4A-C) and found that more than 60 % of mpeg1.1:EGFP-positive cells expressed mCherry-zGem ( Figure 1-figure supplement 4D), confirming that this Tg[EF1α: mCherry-zGem] system works in early macrophages in zebrafish. However, in the retina, the fraction of mCherry-zGem; mpeg1.1:EGFP double-positive cells was less than 2 % of all microglial precursors from 32 to 54 hpf ( Figure 1E). Furthermore, more than 80 % of mpeg1.1:EGFP-positive cells did not incorporate BrdU at 48 hpf ( Figure 1F-H The online version of this article includes the following figure supplement(s) for figure 1: Source data 1. Data for Figure 1C.
Source data 2. Data for Figure 1E.
Source data 3. Data for Figure 1GH.        Embryonic microglial precursor migration into the retina depends on blood vessels The zebrafish retina receives its blood supply from two blood vessel systems, intraocular hyaloid blood vessels encapsulating the lens (Hartsock et al., 2014) and superficial choroidal blood vessels . Developing hyaloid blood vessels start to enter the space between the lens and retina through the ventral fissure at 18-20 hpf. Loop formation occurs around the lens at 24-28 hpf, and a branched hyaloid network forms after 35 hpf (Hartsock et al., 2014). Our live imaging showed that microglial precursors enter the optic cup through the choroid fissure and remain temporarily in the vitreous space between the lens and the retina before they infiltrate the neural retina ( Figure 1-figure supplement 2). Furthermore, microglial precursors start to enter the optic cup after loop formation of hyaloid blood vessels is completed, suggesting a guiding role of blood vessels in microglial precursor colonization of the optic cup. To confirm whether microglial precursors entering the ocular space are associated with developing hyaloid blood vessels, we conducted time-lapse imaging of Tg[kdrl:EGFP; mfap4:tdTomato-CAAX] transgenic embryos, which visualizes endothelial cells of blood vessels (Jin et al., 2005) and ocular microglial precursors, respectively. The first microglial precursor was always associated with ocular hyaloid blood vessels around 30 hpf ( Figure 2A) and moved along blood vessel surfaces (Video 2), so it is very likely that microglial precursors use blood vessels as a scaffold to enter the vitreous space between the lens and the neural retina. Microglial precursors move along hyaloid blood vessels in the ventral fissure, gradually leave vessel surfaces, and invade the neural retina through the basement membrane ( Figure 2B, Figure 2-figure supplement 1, and Video 3). Troponin T2A (tnnt2a; silent heart) is specifically expressed in heart and is essential for heart contraction (Sehnert et al., 2002). In zebrafish brain and mouse retina, hemodynamics drive blood vessel pruning, and loss of blood circulation causes blood vessel regression (Chen et al., 2012;Lobov et al., 2011;Yashiro et al., 2007). To examine whether the entry of microglial precursors into retina is altered upon blood vessel regression, we blocked blood circulation by injecting morpholino antisense oligos against tnnt2a (tnnt2a MO). When blood circulation is inhibited, ocular hyaloid blood vessels do not develop fully and microglial precursors are less likely to be associated with these thin blood vessels ( Figure 2C). The number of ocular microglial precursors was significantly reduced at 36 hpf ( Figure 2D), showing that microglial colonization of the optic cup depends upon normal development of the blood vessel network. This is in contrast to the case of microglial colonization of zebrafish midbrain and optic tectum, which is independent of the blood vessel network (Xu et al., 2016). Indeed, we confirmed that the number of microglial precursors in the optic tectum was not significantly different between tnnt2a morphants and standard MO-injected embryos at 72 hpf, although microglial precursor colonization of the optic tectum was enhanced in tnnt2a morphants at 48 hpf ( Figure 2-figure supplement 2).

Microglial precursors infiltrate the neural retina preferentially through the differentiating neurogenic area
In zebrafish, retinal neurogenesis occurs at the ventronasal retina adjacent to the optic stalk at 25 hpf and propagates into the entire region of the neural retina at 33 hpf (Masai et al., 2000). Microglial precursors start to migrate from the vitreous space into the neural retina after 42 hpf, when the earliest differentiating retinal neurons, RGCs, start to form the IPL (Mumm et al., 2006). To examine the role of retinal neurogenesis and RGC differentiation in microglia precursor infiltration of the neural retina, we used double transgenic lines, Tg[EF1α: mCherry-zGem; mpeg1.1:EGFP], which enable us to examine the relationship between microglial precursor migration and retinal progenitor cells (Mochizuki et al., 2014). Live imaging of Tg[EF1α: mCherry-zGem; mpeg1.1:EGFP] retinas at 42 and 48 hpf  The fraction of microglial precursors that infiltrated mCherry-zGempositive proliferating regions was 7.37 % at 42 hpf and 6.13 % at 48 hpf ( Figure 3C), suggesting that >90% of microglial precursors infiltrate the retina through the mCherry-zGem-negative postmitotic cell region. We also used another transgenic line Tg[ath5:EGFP; mfap4:tdTomato-CAAX]. In the Tg[ath5:EGFP] line, EGFP starts to be expressed in G2 phase of the final neurogenic cell division of retinal progenitor cells and is inherited by their daughter cells, which are negative for BrdU incorporation (Poggi et al., 2005;Yamaguchi et al., 2010), suggesting that ath5:EGFP specifically marks early differentiating retinal neurons. We conducted live imaging of Tg[ath5:EGFP; mfap4:tdTomato-CAAX] retinas at 36, 42, 48 hpf, and found that infiltration of mfap4-positive microglia preferentially occurs in the ath5:EGFP-positive region ( Figure 3D). These data suggest that microglial precursors infiltrate the neural retina preferentially through the neurogenic area, raising the possibility that the neurogenic retinal region acts as a gateway through which microglial precursors move from the vitreous space into the neural retina.
Colonization of the optic tectum by microglial precursors depends on neuronal apoptosis in zebrafish (Casano et al., 2016;Xu et al., 2016). Therefore, it is still possible that microglial precursors preferentially infiltrate the neural retina through the neurogenic region, because of neuronal apoptosis. We inhibited retinal apoptosis by injecting morpholino antisense oligos against p53 (p53 Scale bar: 30 μm. (B) Time-lapse 3D snapshots of Tg[kdrl:EGFP; mfap4:tdTomato-CAAX] eyes for around 3.5 hr after 32 hpf. Ocular microglial precursors and peripheral macrophages outside the optic cup are indicated as yellow-and magenta-colored, surface-rendered objects, respectively, which were prepared from the original scanning image (Figure 2-figure supplement 1). Ocular blood vessels are visualized in green. Microglia associated with hyaloid blood vessels around the lens (white arrows) gradually increase and infiltrate neurogenic retinal tissue (Video 3). Scale bar: 30 μm. (C) Live 3D images of eyes of Tg[kdrl:EGFP; mfap4:tdTomato-CAAX] embryos injected with standard MO and tnnt2a MO. kdrl:EGFP-positive blood vessels (green) are thinner in tnnt2a morphants. Scale bar: 50 μm. (D) Histogram of the number of intraocular microglial precursors in embryos injected with standard MO and tnnt2a MO. Bars and lines indicate means ± SD. ***p < 0.001.
The online version of this article includes the following figure supplement(s) for figure 2: Source data 1. Data for Figure 2D.

Neurogenesis acts as a gateway for microglial precursors to enter the retina
To confirm the possibility that the neurogenic retinal region functions as a gateway for microglial precursors to infiltrate the retina, we examined whether microglial precursor migration into the retina is compromised when retinal neurogenesis is affected. Previously, we found that histone genesis slowed in zebrafish stem loop binding protein 1 (slbp1) mutants, leading to severe delays in retinal neurogenesis (Imai et al., 2014). Our bulk RNAseq analysis confirmed that retinal neurogenesis and subsequent neuronal differentiation were markedly delayed in zebrafish slbp1 mutants (Figure 4figure supplement 1), such that ath5 expression spread into the entire slbp1 mutant retina only at 48 hpf, an event that occurs in wild-type retina at 33 hpf (Imai et al., 2014). We combined slbp1 mutants with transgenic lines Tg[ath5:EGFP; mfap4: tdTomato-CAAX] and examined the number of ocular mfap4:tdTomato-CAAX-positive microglial precursors ( Figure 4A, Figure 4-figure supplements 2A and 3). In 48-hpf slbp1 mutant retinas, the number of ocular microglial precursors was 4.67 ± 2.42 ( Figure 4A and B), which is similar to the number in wild-type retinas at 32 hpf ( Figure 1B), whereas the number of ocular microglial precursors in wild-type siblings was 17.60 ± 5.13 at 48 hpf ( Figure 4A and B). To confirm whether the slbp1 mutation interferes with genesis of early macrophages, we examined peripheral mfap4 + cells in the tail/trunk region of slbp1 mutants and wild-type sibling embryos. There was no significant difference in mfap4 + cells between slbp1 mutants and wild-type siblings in the trunk/tail region ( Figure 4C and D), indicating that the slbp1 mutation does not influence early macrophage specification in zebrafish embryos. Although inhibition of retinal apoptosis by p53 MO does not influence microglial precursor colonization of the retina (Figure 3-figure supplement 3A-B ), we examined the level of retinal apoptosis in slbp1 mutants. TUNEL revealed that apoptosis was increased in slbp1 mutant retinas compared with wild-type sibling retinas (Figure 4-figure  supplement 4). These data exclude the possibility that decreased retinal apoptosis affects microglial precursor colonization of the retina in slbp1 mutants, and again confirm that neuronal apoptosis is not the major cue for microglial precursor colonization of the retina. Mouse brain cortex colonization by microglia depends on the Cxcl12a-CxcR4 signaling axis (Arnò et al., 2014). We previously reported that cxcl12a expression is absent in the optic stalk of zebrafish slbp1 mutants (Imai et al., 2014). To exclude the possibility that the absence of cxcl12a expression in the optic stalk affects microglial colonization of the retina in zebrafish slbp1 mutants, we examined associated with the mCherry-zGem-positive area (black) and the mCherry-zGem-negative area (grey). The fraction of microglial precursors associated with the mCherry-zGem-positive area is only 7.37 % at 42 hpf and 6.13 % at 48 hpf. Thus, more than 90 % of microglial precursors are located in the mCherry-zGem-negative retinal area. (D) Live images of Tg[ath5:EGFP; mfap4:tdTomato-CAAX] retinas at 36 (upper panels), 42 (middle panels) and 48 hpf (bottom panels). Three confocal scanning plane levels are indicated as superficial (a'-a"'), middle (b'-b"'), and deep (c-c"'). Dotted circles indicate the outline of the optic cup. The right-most column images indicate higher magnification images shown in the square of left panels. mfap4-positive microglia (magenta, arrows) are closely associated with ath5-positive neurogenic cells (green). Scale bar: 50 μm, except the right-most column images (Scale bar: 15 μm).
The online version of this article includes the following figure supplement(s) for figure 3: Source data 1. Data for Figure 3C.        zebrafish cxcl12a morphants. Injection of cxcl12a-MO at 500 μM, which effectively induces RGC axon trajectory defects reported in zebrafish odysseys mutants carrying mutations in cxcl12a receptor, cxcr4b , did not affect the number of ocular microglial precursors (Figure 4-figure  supplement 5). Thus, Cxcl12a-CxcR4 signaling is not involved in microglial colonization defects in slbp1 mutants. We also confirmed that elimination of microglial precursors with pu.1 MO did not affect the rate of retinal neurogenesis or cell differentiation by 72 hpf (Figure 4-figure supplement  6). We previously showed that overexpression of Notch1 intracellular domain (NICD) suppresses retinal neurogenesis in zebrafish . We confirmed that overexpression of NICD suppresses retinal neurogenesis in zebrafish by injecting a DNA expression construct encoding UAS:myc-NICD (Scheer and Campos-Ortega, 1999) into Tg[hsp:gal4; ath5:EGFP] double transgenic embryos (Figure 4-figure supplement 7). Next, we examined whether microglial precursor infiltration of the retina is compromised in retinas overexpressing NICD. We established a zebrafish transgenic line, Tg[rx1:gal4-VP16], which expresses Gal4-VP16 under control of a retinal progenitor-specific promoter rx1 (Chuang et al., 1999), and then injected two DNA expression constructs encoding UAS:EGFP (Köster and Fraser, 2001) and UAS:myc-NICD into Tg[rx1:gal4-VP16; mfap4:tdTomato-CAAX] double-transgenic embryos. Embryos injected with only the DNA construct of UAS:EGFP served as a positive control. We selected embryos in which EGFP was expressed in most retinal cells at 24 hpf and used them for further analysis. The number of ocular microglial precursors was significantly reduced in embryos overexpressing NICD and EGFP, compared with control embryos overexpressing EGFP, at 44 hpf ( Figure 4E and F, Figure 4-figure supplements 2B and 3). These data support Source data 1. Data for Figure 4B.
Source data 2. Data for Figure 4D.
Source data 3. Data for Figure 4F.
Source data 4. Data for Figure 4H.            The blockade of retinal neurogenesis delays differentiation of the first-born retinal cell-type, RGCs. To examine whether blockade of RGC differentiation affects microglial precursor colonization of the neural retina, we applied an antisense morpholino against ath5 (known as atoh7) (ath5 MO). As with the zebrafish ath5 mutant, lakritz (Kay et al., 2001), RGC differentiation was specifically inhibited in ath5 morphant retinas (Figure 4-figure supplement 8A, B). In ath5 morphants, the timing of the first appearance of microglial precursors in the ocular vitreous space was not altered, but the number of ocular microglial precursors was significantly decreased at 49 hpf ( Figure 4G and H, and Figure 4figure supplements 2C and 3 and 8 C), suggesting that RGC differentiation or RGC-mediated IPL formation is required for microglial precursor infiltration into the neural retina.

Microglial precursors preferentially associate with neurogenic retinal columns
To determine whether microglia precursors have greater affinity for differentiating neurons than for retinal progenitor cells, we carried out two sets of experiments. First, we conducted cell transplant experiments using a wild-type donor line and an slbp1 mutant recipient line carrying Tg[mfap4:tdTomato-CAAX]. Wild-type donor cells were transplanted into slbp1 mutant recipient embryos at the blastula stage. We selected slbp1 mutant and wild-type sibling embryos in which wild-type, donor retinal cell columns were introduced in a mosaic manner at 48 hpf ( Figure 5A). Host microglial precursors and donor retinal cells were visualized with mfap4:tdTomato-CAAX and Alexa-488 Dextran, respectively. In slbp1 mutant host retinas, microglial precursors were likely to be associated with donor wild-type retinal columns more frequently than in wild-type host retinas ( Figure 5B). To analyze these data statistically, we compared eyes in which wild-type donors were transplanted into wild-type hosts with those in which wild-type donors were transplanted into slbp1 mutant hosts ( Figure 5-figure supplement 1A-B). The fraction of microglial precursors associated with donor wild-type retinal columns in total ocular microglial precursors was significantly higher in slbp1 mutant host retinas than in wild-type sibling host retinas at 48 hpf ( Figure 5C), suggesting that microglial precursors are more attracted by wild-type donor neurogenic retinal columns than surrounding slbp1 mutant proliferative retinal cells. Since the fraction of microglial precursors associated with donor retinal columns in total microglial precursors may depend on the number of donor retinal columns incorporated into host retinas, we next estimated trapping efficiency of microglial precursors per donor column by dividing the fraction of microglial precursors associated with donor columns with the transplanted donor column number in each eye ( Figure 5-figure supplement 1B). Trapping efficiency of microglial precursors per donor column was significantly higher in slbp1 mutant host retinas than in wild-type sibling host retinas ( Figure 5D), suggesting that microglial precursors are preferentially associated with donor-derived wild-type retinal cells than with host-derived slbp1 mutant retinal cells.
Second, we injected two DNA constructs encoding UAS:EGFP and UAS:mycNICD into Tg[hsp:gal4; mfap4:tdTomato-CAAX] double-transgenic wild-type embryos. Two rounds of heat-shock treatment at 18 and 30 hpf induced expression of NICD and EGFP in a mosaic manner in the retina ( Figure 5E). We examined the fraction of mfap4:tdTomato-CAAX-positive microglial precursors associated with EGFP-expressing retinal columns in the total number of mfap4:tdTomato-CAAX-positive microglial precursors ( Figure 5F). This fraction was significantly lower in retinas overexpressing NICD and EGFP than in control retinas overexpressing only EGFP ( Figure 5G, Figure 5-figure supplement 1C). We also confirmed that trapping efficiency of mfap4:tdTomato-CAAX-positive microglial precursors per EGFP-positive retinal column was significantly lower in retinas overexpressing NICD and EGFP than in retinas overexpressing only EGFP ( Figure 5H, Figure 5-figure supplement 1C). Thus, microglial precursors are less attracted by retinal columns in which neurogenesis is arrested. Taken together, these data suggest that microglial precursors preferentially associate with neurogenic retinal columns as opposed to proliferative retinal columns.

IL34 is involved in microglial precursor colonization of the retina
Recently, it was reported that microglial colonization of zebrafish brain, including retina, depends on CSF-R, and that one of the CSF-R ligands, IL34, dominates this process (Wu et al., 2018). In adult mouse retina, RGCs express IL34, which attracts one subset of microglia and retains them around the  et al., 2019). First, we confirmed that retinal cell differentiation proceeds normally until 72 hpf in zebrafish il34 mutants, although pyknotic nuclei were stochastically observed in RGC and amacrine cell layers (Figure 6-figure supplement 1). Next, we examined microglial precursor colonization of the retina. The number of ocular microglial precursors was significantly lower in il34 homozygous mutants than in wild-type siblings at 34 hpf ( Figure 6-figure supplement 2) and 48 hpf ( Figure 6A and B). Thus, consistent with the previous report (Wu et al., 2018), IL34 is required for microglial precursor colonization of the retina in zebrafish. However, il34 mRNA expression is comparable in slbp1 mutant heads and wild-type sibling heads at 48 hpf ( Figure 6-figure supplement 3), suggesting that il34 mRNA expression is not linked to retinal neurogenesis. Since the number of ocular microglial precursors in il34 homozygous mutants was zero at 34 hpf ( Figure 6-figure supplement 2) and no more than two, if any, at 48 hpf ( Figure 6B), it is likely that Csf1r-il34 signaling promotes microglial precursor movement from yolk to the optic cup upstream of the blood-vesselmediated guidance mechanism ( Figure 6C).

Discussion
In zebrafish, primitive microglia originate from the RBI, which is a hematopoietic tissue equivalent to mouse yolk sac, whereas definitive microglia are generated from hematopoietic stem cells that are specified in the VDA (Ferrero et al., 2018;Xu et al., 2015). Primitive and definitive waves of hematopoiesis generate embryonic and adult microglia, respectively. Using zebrafish as an animal model, several groups investigated microglial colonization from the periphery into developing brain, especially the optic tectum, which is part of the midbrain (Casano et al., 2016;Herbomel et al., 2001;Svahn et al., 2013;Wu et al., 2018;Xu et al., 2016). Colonization of the optic tectum by microglial precursors depends on neuronal apoptosis, probably through attraction by an apoptotic cell-secreted phospholipid, lysophosphatidylcholine (LPC) (Casano et al., 2016;Xu et al., 2016). In addition, microglial colonization of brain is CSF receptor-dependent (Herbomel et al., 2001;Wu et al., 2018). In mice, microglial colonization of brain requires functional blood circulation (Ginhoux et al., 2010). However, in zebrafish, microglial colonization of the optic tectum is independent of blood circulation (Xu et al., 2016). A series of elegant studies revealed the molecular network that promotes microglial colonization of the midbrain. However, it remains to be seen whether this mechanism fully explains colonization of other brain regions by microglial precursors. In this study, we focused on zebrafish retina and investigated the mechanism that regulates migration of embryonic microglial precursors into developing retina.
We first conducted live imaging of zebrafish microglial precursors from 24 to 54 hpf. Microglial precursors progressively increase in number during embryonic development. Interestingly, almost all microglial precursors enter the optic cup through the choroid fissure. However, peripheral macrophages located in the mesenchymal region between the eye and the brain did not enter the optic cup across the ciliary marginal zone. This may be consistent with the observation that these peripheral macrophages never enter the retina following rod cell death , suggesting a functional difference between peripheral macrophages and ocular microglia. Next, we found that the majority of ocular microglial precursors do not undergo S phase and are probably in G1 phase. Thus, the increase of ocular microglial precursors is due to migration from outside the eye. In developing mouse retina, microglial precursors appear from the vitreous area near the optic disk at E11.5, progressively increase in number, and then infiltrate the neural retina. These retinal microglia were retinal cell columns versus the total number of microglial precursors in the optic cup. The average fraction of mfap4-positive cells associated with EGFP-positive retinal columns is significantly lower in retinas injected with UAS:EGFP+ UAS:myc-NICD than with only UAS:EGFP control. Bars and lines indicate means ± SD. ***p < 0.005. (H) The trapping efficiency of mfap4-positive microglial precursors per EGFP-expressing retinal cell columns. The average trapping efficiency is significantly lower in retinas injected with UAS:EGFP+ UAS:myc-NICD than with only UAS:EGFP control, suggesting less affinity of microglial precursors for proliferative NICD-expressing retinal cells. Bars and lines indicate means ± SD. *p < 0.05.
The online version of this article includes the following figure supplement(s) for figure 5: Source data 1. Data for Figure 5CDGH.  At 48 hpf, iridophores start to differentiate around the optic cup, which causes a noise signal (magenta) in confocal scanning. Using the surface-rendering tool of Imaris software (Bitplane), we eliminated iridophore-derived noise and extracted mfap4:tdTomato-CAAX signals from ocular microglial precursors (green) (see the legend of Figure 4-figure supplement 3). Scale Figure 6 continued on next page also negative for a proliferative marker, Ki67 (Santos et al., 2008), suggesting that mouse embryonic retinal microglia are also non-proliferative.
Another interesting finding is that entry of microglial precursors into the optic cup through the choroid fissure depends on ocular blood vessels. We observed that migrating microglial precursors are closely associated with hyaloid blood vessels after loop formation. These microglial precursors pass along these vessels, which traverse the choroid fissure and surround the posterior region of the lens. Furthermore, the number of ocular microglial precursors was reduced when blood circulation was blocked. Since inhibition of blood circulation compromises the structural integrity of blood vessels in zebrafish, we conclude that ocular blood vessel formation is required for microglial precursor entry into the optic cup through the choroid fissure. One possibility is that blood vessels function as a path upon which microglial precursors enter the optic cup. Membrane proteins or extracellular matrix proteins on blood endothelial cells may facilitate the association of microglial precursors with blood vessel surfaces. Alternatively, substances that attract microglial precursors may be released from hyaloid blood endothelial cells. Previous studies on human and murine microglia demonstrated that microglial colonization of the retina takes place prior to retinal vascularization, and that microglia facilitate ocular blood vessel development (Checchin et al., 2006;Fantin et al., 2010;Rymo et al., 2011). Macrophages initiate endothelial cell death for blood vessel regression in developing mouse retina (Lang and Bishop, 1993;Lobov et al., 2005). However, in contrast to mammals, elimination of microglia by pu.1 MO or irf8 mutation did not affect ocular blood vessel formation in zebrafish, suggesting that microglia do not regulate ocular blood vessel formation in zebrafish. Interestingly, classic histological studies on mouse retinas showed that early emerging ocular microglia are associated with the hyaloid artery (Hume et al., 1983;Santos et al., 2008), which is located in the vitreous area and regresses in later stages before retinal vasculature formation (Ito and Yoshioka, 1999). Thus, further investigation will be necessary to determine whether the hyaloid artery guides microglial precursors into the optic cup in vertebrate species such as mice. In zebrafish, colonization of the optic tectum by microglia is independent of blood circulation (Xu et al., 2016). We confirmed that the number of microglia in the optic tectum did not differ between tnnt2a morphants and control embryos at 72 hpf; however, microglial colonization of the optic tectum was enhanced and microglial shape was round rather ramified in tnnt2a morphants at 48 hpf. Further study will be necessary to clarify the role of blood circulation in microglial colonization of the optic tectum.
After 42 hpf, microglial precursors detach from hyaloid blood vessels and start to infiltrate the neural retina. Interestingly, we found that more than 90 % of microglial precursors enter the neural retina through the neurogenic area. Indeed, the number of microglial precursors is reduced in slbp1 mutant retinas and NICD-overexpressing retinas, in both of which retinal neurogenesis is severely delayed. Furthermore, we conducted two sets of experiments: the first was cell transplantation from wild-type donor cells into slbp1 mutant host retinas, which introduced neurogenic wild-type retinal cell columns in proliferative slbp1 mutant retinas, and the second was overexpression of NICD in wildtype retina, which introduced proliferative retinal cell columns in neurogenic retinas. Consistently, in both cases, microglial precursors were preferentially associated with neurogenic retinal cell columns.
bar: 50 μm. (B) Histogram of numbers of ocular microglial precursors in wild-type, il34 heterozygous and homozygous mutant retinas at 48 hpf. The number of ocular microglial precursors is almost zero, and very few, if any (one or two), in il34 homozygous mutants, indicating that ocular microglial precursors are significantly reduced in il34 homozygous mutants. The number of ocular microglial precursors is mildly reduced in il34 heterozygous mutants, but does not differ significantly from that of wild-type siblings. Bars and lines indicate means ± SD. ***p < 0.005. (C) A possible model of the guidance mechanism of microglial precursor into zebrafish retina. IL34 is involved in movement of microglial precursors toward the brain. Microglial precursors continue into the optic cup along blood vessels, and subsequently infiltrate the neural retina through the neurogenic area.
The online version of this article includes the following figure supplement(s) for figure 6: Source data 1. Data for Figure 6B.

Figure 6 continued
Thus, neurogenesis is required for infiltration of microglial precursors into the neural retina after 42 hpf. We observed that the number of microglial precursors is diminished in ath5 morphant retinas, suggesting that RGCs are required for infiltration of microglial precursors into the neural retina. There are at least three possible mechanisms for this infiltration. First, the basal region of retinal progenitor cells may function as a physical barrier that inhibits microglial precursor infiltration of the neural retina. Second, microglial precursors may be attracted to surfaces of differentiating retinal neurons or RGCs. Third, differentiating retinal neurons or RGCs may release a specific attractant for microglia. There are several candidate molecules that suggest the third possibility. In adult mice, RGCs express IL34, which attracts microglia and retains them around the IPL niche (O'Koren et al., 2019). Indeed, microglial colonization of zebrafish brain depends on CSF-R, and one of the CSF-R ligands, IL34, dominates this process (Wu et al., 2018). We confirmed that microglial precursor colonization of retina is severely affected in il34 mutants. However, il34 mRNA expression is comparable in slbp1 mutants and their wild-type siblings, suggesting that IL34 is not linked to neurogenesis-mediated microglial precursor infiltration. Rather, the number of ocular microglial precursors in il34 mutants was almost zero at 34 and 48 hpf, so it is very likely that Csf1r-il34 signaling initiates microglial precursor movement from yolk toward brain and retina, followed by blood vessel-and neurogenesis-mediated guidance.
It was reported that apoptosis attracts microglia in zebrafish developing brain (Casano et al., 2016;Xu et al., 2016). However, microglial precursor colonization of the retina is normal in zebrafish p53 morphants, suggesting that apoptosis does not promote microglial precursor colonization of the retina. Why are microglial precursors insensitive to retinal apoptosis? We found that apoptosis is enhanced in zebrafish slbp1 mutant retinas, in which microglial precursor colonization is severely affected due to a delay of retinal neurogenesis. It is likely that spontaneous apoptotic cells fail to be eliminated because of the reduced number of microglial precursors in slbp1 mutant retinas; however, interestingly, these increased dead cells did not promote microglial precursor infiltration into slbp1 mutant retinas, suggesting that neurogenesis primarily opens the gate through which microglial precursors enter the neural retina. Since retinal neurogenesis normally occurs from 24 to 48 hpf in zebrafish, microglial precursors could not be attracted by apoptosis without the infiltration path opened by neurogenesis before 48 hpf. Further studies will be necessary to unveil the molecular mechanism underlying microglial infiltration into neural retina.
In summary, there are three mechanisms for microglial colonization of developing zebrafish retina ( Figure 6C). IL34-CSF-R signaling initiates microglial precursor movement from yolk toward brain and retina. Microglial precursors further use ocular hyaloid blood vessels as a pathway to enter the optic cup and then infiltrate the neural retina preferentially through the neurogenic region. In the future, it remains to identify molecules involved in blood-vessel-and neurogenesis-mediated guidance mechanisms, and to assess whether these mechanisms are used for microglial colonization of other brain regions in other vertebrate species.

Morpholino
Morpholino antisense oligos were designed as shown below.

Cell transplantation
Cell transplantation was performed as previously described (Masai et al., 2003). Wild-type zygotes were injected with Alexa-488 dextran (Thermo Fisher Scientific, D22910) and used for donor embryos. slbp1 mutant embryos carrying Tg[mfap4:tdTomato-CAAX] were used as host embryos. Host embryos carrying donor retinal cells were selected by observing Alexa 488 fluorescence at 24 hpf. slbp1 mutant and wild-type sibling embryos were sorted based on the slbp1 mutant morphological phenotype at 48 hpf and used for live imaging. After confocal images were obtained, the number of ocular mfap4-positive microglial precursors associated with Alexa-488 dextran-labeled donor transplanted retinal columns was counted. The fraction of ocular mfap4-positive microglial precursors associated with donor transplanted retinal columns in total ocular microglial precursors was calculated. The trapping efficiency of ocular mfap4-positive microglial precursors per transplanted donor retinal column was calculated using the total number of donor transplanted retinal columns in the retina. Detailed information on each transplanted eye is shown in Figure 5-figure supplement 1A-B.

RNA extraction
Heads of 48 hpf wild-type sibling and slbp1 mutant embryos were dissected and transferred to 100 μL Sepasol (Nacalai tesque, 09379) on ice. Heads were then homogenized using a hand homogenizer (~20 pulses). Twenty μL CHCl 3 were then added to samples and mixed gently. After centrifugation at 15,000 g for 15 min, the aqueous phase was collected and mixed with 100 μL isopropanol. One μL of RNase-free glycogen (Nacalai tesque 11170-11, 20 mg/mL) was added to all samples to increase the yield. After incubating at room temperature for 10 min, samples were centrifuged at 15,000 g at 4 °C for 15 min. Supernatant was removed and the pellet was washed three times with 500 μL of 75 % ethanol and centrifuged at 8000 g at 4 °C. The pellet was then resuspended in a desired amount of nuclease-free water and stored at -80 °C. RNA concentration and purity of samples were determined using a Nanodrop.

RNA sequencing and analysis
RNA samples with RIN >7 were subjected to paired-end sequencing using an Illumina HiSeq4000. First, a quality check was performed using FastQC and read trimming was done with Trimomatic (Bolger et al., 2014). PRINSEQ lite (Schmieder and Edwards, 2011) was used for PolyA trimming and quality filtering. Trimmed sequences were then mapped to the zebrafish reference genome (GRCz11) using hisat2.1.0 (Kim et al., 2019) and mapped reads are counted using featureCounts (Liao et al., 2014). With the R package, EdgeR (Robinson et al., 2010), differentially expressed genes with Log 2 FC > |2| and FDR values < 0.01 were extracted. EnhancedVolcano package (Blighe et al., 2018) was used to draw volcano plots. A heat map was generated with the pheatmap package (Kolde, 2019).

Evaluation of Il34 mRNA expression by semi-quantitative PCR
Extracted RNA from 48-hpf wild-type sibling and slbp1 mutant heads was used to prepare cDNA, using ReverTra Ace qPCR RT master mix with gDNA remover (Toyobo, FSQ-301). The expression level of il34 mRNA was evaluated with quantitative PCR using the primers below. mRNA of cytoplasmic actin β2, namely actb2 (ZFIN), was used for normalization.

Mosaic expression of NICD in retinal cells using Tg[rx1:gal4-VP16] and Tg[hsp:gal4] transgenic lines
The DNA fragment that covers a 2892 bp genomic region upstream from the start codon of rx1 cDNA (Chuang et al., 1999), was amplified by PCR and inserted between XhoI and BamHI sites of the Tol2 base expression vector, pT2AL200R150G (Urasaki et al., 2006). Next, DNA fragments encoding gal4-VP16 (Köster and Fraser, 2001) were further inserted between BamHI and ClaI sites of pT2AL200R150G to fuse the rx1 promoter. The plasmid was injected with Tol2 transposase mRNA into fertilized eggs of the UAS:EGFP transgenic line to establish a transgenic line, Tg[rx1:gal4-VP16] oki065 . A mixture of plasmids of UAS:EGFP (Köster and Fraser, 2001) and UAS:myc-NICD (Scheer and Campos-Ortega, 1999) (each 10 ng/μL) were injected into fertilized eggs of the Tg[mfap4:tdTomato-CAAX; rx1:gal4-VP16] or Tg[mfap4:tdTomato-CAAX; hsp:gal4] transgenic line. In the case of the Tg[mfap4:tdTomato-CAAX; hsp:gal4] transgenic line, two rounds of heat shock at 37 °C for 1 hr were applied at 18 and 30 hpf. Embryos expressing EGFP in the optic cup were selected at 24 hpf, fixed with PFA at 48 hpf and used to prepare serial retinal sections for imaging analysis. After confocal images were obtained, the number of ocular mfap4-positive microglial precursors associated with EGFP-expressing columns was counted. The fraction of ocular mfap4-positive microglial precursors associated with EGFP-expressing columns in total microglial precursors was calculated and the trapping efficiency of ocular mfap4-positive microglial precursors per EGFP-expressing column was calculated using the total number of EGFP-expressing columns in the retina. Detailed information on each injected eye is shown in Figure 5-figure supplement 1C. To confirm that NICD inhibits retinal neurogenesis, UAS:myc-NICD or UAS:mCherry (each 10 ng/μL) was injected into zebrafish transgenic embryos Tg[ath5:EGFP;hsp:gal4]. Three rounds of heat shock at 37 °C for 1 hr were applied at 18, 24, and 30 hpf. Embryos were fixed at 36 hpf and labeled with anti-myc tag antibody to visualize myc-NICD expressing retinal cells with Alexa-543-conjugated secondary antibody. Whole retinas were used for confocal scanning with an FV3000RS (Olympus). Controls were UAS:mCherry-injected samples and used directly for live confocal scanning. Confocal 3D retinal images were used to count the number of ath5:EGFP-positive and negative retinal columns in myc-NICD or mCherry expressing retinal columns from five independent embryos.

Evaluation of microglial precursor colonization of the retina in Il34 mutants
The il34 hkz11 allele (Wu et al., 2018) was combined with the Tg[mfap4:tdTomato] transgenic line and used for analysis. Embryos were generated by pair-wise crosses between heterozygous mutant male and female fish, and maintained with N-phenyl thiourea (PTU)-containing water to prevent melanophore pigmentation. Whole retinas of 19 embryo at 34 hpf and 29 embryos at 48 hpf were scanned with confocal microscopy, using an LSM710 (Zeiss) or an FV3000RS (Olympus). Embryos were fixed with 4 % PFA and used for genotyping. A DNA fragment containing the 4-base deletion mutation of the il34 hkz11 allele was amplified by PCR and sequenced to determine genotypes. Primers used for PCR amplification and sequencing are below.
Using the surface rendering tool of Imaris software (Bitplane, ver.9.1.2), we eliminated signals of iridophore-derived noise or peripheral macrophages around the optic cup and extracted only ocular microglial precursors. The number of ocular microglial precursors was counted in each retina and compared between genotypes.

Statistical analysis
Statistical analyses were performed using GraphPad Prism version 8.2.1. Statistical significance was determined using two-tailed unpaired Student's t-tests for Figure 1G; Figure 2D; Figure

Funder
Grant reference number Author The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.