A Dynamic Change of Microglial States Occurs During the Transition From Photoreceptor Degeneration to Regeneration in Zebrafish pde6c Mutants

ABSTRACT

Microglia, brain‐resident immune cells, maintain brain homeostasis. However, homeostatic microglia dynamically change into disease‐associated microglia in human neurodegenerative diseases. Furthermore, microglia‐mediated inflammation is required to initiate neuronal regeneration in zebrafish brain. To understand how functional states of microglia are regulated in response to neuronal degeneration and regeneration, we focused on pde6c mutants, a chronic photoreceptor degeneration zebrafish model. We conducted scRNA‐seq analysis on microglia in wild‐type sibling and pde6c mutant retinas at the onset of photoreceptor degeneration (5 dpf) and Müller glia‐mediated neuronal regeneration (4 wpf). At 5 dpf, retinal microglia consist of three clusters, which correspond to homeostatic, degeneration‐response, and stress‐response microglia, respectively. The degeneration‐response cluster expands in pde6c mutants and expresses genes for neuroprotection and tissue repair. At 4 wpf, retinal microglia comprise four clusters, two of which are specifically produced in pde6c mutants and approach the photoreceptor layer. Furthermore, another cluster is prominently localized in the retinal stem cell niche and shows a transcriptomic profile similar to that of neurogenic‐associated microglia (NAM). Comparison of transcriptomic similarity between 4 wpf and 5 dpf microglial clusters revealed that each 4 wpf microglial cluster inherits characteristics of homeostatic, degeneration‐response, and stress‐response state of 5 dpf microglia in different combinations. Thus, there is a unique heterogeneity of microglia in the initial stage of Müller glia‐mediated regeneration. Taken together, our findings reveal a dynamic change of retinal microglia during the transition from photoreceptor degeneration to Müller glia‐mediated neuronal regeneration in zebrafish.

Zebrafish pde6c mutants undergo photoreceptor degeneration from 5 dpf, but rods regenerate from 4 wpf. scRNA‐seq of retinal microglia reveals dynamic change of states between 5 dpf and 4 wpf. Two states emerge specifically during regeneration.

1. Introduction

Microglia are brain‐resident immune cells and maintain healthy conditions in the brain by clearing apoptotic cells through phagocytosis to prevent inflammation (Bohlen et al. 2019). Recent single‐cell RNA sequencing (scRNA‐seq) analysis revealed that homeostatic microglia, whose signatures are normally specified by TGF‐β signaling (Butovsky et al. 2014), dynamically change into a unique set of microglia called disease‐associated microglia (DAM) during the pathological process in Alzheimer model mice (Butovsky and Weiner 2018; Keren‐Shaul et al. 2017). Further scRNA‐seq studies on human patients and mouse neurodegenerative disease models revealed that heterogeneity and plasticity of microglia appear under neurodegenerative conditions in a context‐dependent manner, varying with brain area, disease type, and aging (Hammond et al. 2019; Lan et al. 2024; Masuda et al. 2019; Sankowski et al. 2019). Therefore, it is important to understand how microglial functional states are regulated during pathogenesis of neuronal degenerative diseases.

In contrast to mammals, zebrafish show remarkably high regenerative potential in the nervous system (Grandel and Brand 2013; Lindsey and Tropepe 2006). Injuries to the zebrafish telencephalon cause an acute inflammatory response mediated by microglia; however, persistent gliosis, scar formation, and chronic inflammation do not occur (Baumgart et al. 2012; Kishimoto et al. 2012; Kroehne et al. 2011). Rather, radial glia are reprogrammed to become neural stem cells and to rebuild neural circuits of damaged tissue. Interestingly, in zebrafish, an acute inflammatory response after brain injury is required to initiate the regenerative response of radial glia in the telencephalon (Kyritsis et al. 2012), suggesting a novel role of microglia in neuronal regeneration. In addition to roles in neuronal degeneration and regeneration, new subtypes of microglia associated with neurogenic regions or synaptic regions were identified in the optic tectum and midbrain of zebrafish, and they are denominated “neurogenic‐associated microglia” (NAM) and “synaptic region‐associated microglia” (SAM), respectively (Silva et al. 2021), indicating specialized roles of microglia in normal homeostatic brain functions.

Vertebrate retina consists of six major classes of neurons, namely retinal ganglion cells (RGCs), three interneurons (amacrine cells, bipolar cells and horizontal cells), and two types of photoreceptors (rods and cones), which form the RGC layer (RGCL), the inner nuclear layer (INL) and the outer nuclear layer (ONL), respectively (Figure 1A). The two synaptic layers, namely the inner plexiform layer (IPL) and the outer plexiform layer (OPL), form at the interface between the RGCL and INL, and between the INL and ONL, respectively. The most apical region of photoreceptors develops to form a multi‐stacked photoreceptive membrane structure called the outer segment (OS), which is positioned beneath the retinal pigment epithelium (RPE). During retinal development, multipotent retinal progenitor cells generate these retinal neurons as well as a type of glial cell called Müller glia. After zebrafish retinas suffer damage, Müller glia initiate cell proliferation to generate a pool of neural progenitor cells, which differentiate into retinal neurons and regenerate damaged neural circuits (Lahne et al. 2020). Molecular mechanisms underlying neuronal regeneration have been investigated (Jui and Goldman 2024) to identify key regulators, Achaete‐scute complex‐like 1a (Ascl1a) (Fausett et al. 2008) and Heparin‐binding epidermal‐like growth factor (hb‐Egf) (Wan et al. 2012). Downstream of Ascl1a, Insulinoma‐associated 1a (Insm1a) regulates later steps of Müller glia‐mediated neuronal regeneration (Ramachandran et al. 2012). Pax6a/b and Sox2 promote Müller glia reprogramming (Gorsuch et al. 2017; Thummel et al. 2010). Matrix metalloproteinase 9 (Mmp‐9) suppresses Müller glia‐derived progenitor proliferation, and prompts survival of regenerating photoreceptors (Silva et al. 2020). However, it is unknown how microglia modulate the Müller glia‐mediated regeneration program.

FIGURE 1.

Microglia phagocytose dying photoreceptors in zebrafish pde6c mutants at 5 dpf. (A) Schematic drawing of zebrafish retinal neural circuit. RGCL, retinal ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear layer; OS, photoreceptor outer segment; RPE, retinal pigmented epithelium. (B) Pathological process of photoreceptor degeneration and Müller glia‐mediated rod regeneration in zebrafish pde6c mutants. (C) Retinal sections of wild‐type sibling and pde6c mutant embryos at 3–5 dpf. Microglia and nuclei are visualized with Tg[mpeg1.1:EGFP] (green) and Hoechst staining (gray), respectively. Apoptotic cells are labeled with TUNEL (magenta). Upper panels indicate only images of green and magenta channels. Dotted lines indicate the OPL. Scale bars: 20 μm. (D) Number of TUNEL‐positive ONL cells per section. Each dot represents the average of 4 sections from one fish. Nested t‐tests (two tailed): means ± SD, ns, p > 0.05; ***p ≤ 0.001. (E) Number of microglia in the ONL per section. Each dot indicates the average of 4 sections from one fish. Nested t‐tests (two tailed): means ± SD, ns, p > 0.05; **p ≤ 0.01. (F) Projection view of 3D confocal scanning of 5 dpf wild‐type sibling and pde6c mutant retinas. Microglia are visualized with Tg[mpeg1.1:EGFP] (green). Dotted lines indicate the outline of the retina. The right‐hand graph indicates the number of microglia per eye. Unpaired t‐tests (two tailed): means ± SD, **p ≤ 0.01. Scale bars: 30 μm. (G) Live imaging of 5 dpf wild‐type sibling and pde6c mutant retinas carrying the Tg[mfap4:tdTomato‐CAAX] transgene (green). Dying cells are stained with acridine orange (magenta). Inset panels show higher magnification images of the ONL shown in the top panels (dotted rectangle). The third row represents IMARIS rendering of acridine orange signals as spots and mfap4:tdTomato signal as surface objects. The bottom panel shows the object‐rendered images filtered to show only acridine orange‐derived spots engulfed by microglia (yellow arrowheads). Dotted lines indicate the OPL. Scale bars: 30 μm. (H) qPCR of microglia‐related genes in wild‐type sibling and pde6c mutant eyes at 5 dpf. Three biological replicates of 20 eyes each were tested. Unpaired t‐test: means ± SD. ns, p > 0.05; *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

The role of microglia in Müller glia‐mediated regeneration in zebrafish has been studied using acute damage induced by application of a neurotoxin (ouabain), light exposure, N‐Methyl‐D‐aspartic acid (NMDA) treatment, and stab injury. In zebrafish, microglia rapidly accumulate at sites of neuronal damage induced by ouabain; however, they remain amoeboid in shape and localized to the damage site even as retinal regeneration begins, suggesting important roles of microglia in neuronal degeneration and regeneration (Mitchell et al. 2018). Dexamethasone treatment, which inhibits inflammation, suppresses Müller glia‐mediated regeneration induced by either light exposure (Silva et al. 2020) or NMDA treatment (Iribarne and Hyde 2022). Furthermore, Müller glia‐mediated regeneration induced by stab injury is suppressed in zebrafish transgenic line Tg[mpeg1: nitroreductase (NTR)‐mCherry] treated with metronidazole (MTZ), which eliminates microglia (Zhang et al. 2020). On the other hand, zymosan A, which enhances inflammation, promotes Müller glia‐mediated regeneration induced by stab injury (Zhang et al. 2020). Thus, inflammation is required for Müller glia‐mediated regeneration in response to acute damage. Consistently, Müller glia‐mediated regeneration in response to acute light damage was suppressed in zebrafish interleukin 1β (il1b) mutants (Lu and Hyde 2024). Interestingly, an anti‐inflammatory cytokine, interleukin 10 (il10), which is normally elevated following transient activation of il1b after acute damage, is also required for Müller glia‐mediated regeneration (Lu and Hyde 2024). Since both il1b and il10 are expressed in microglia, it is likely that pro‐ and anti‐inflammatory microglia cooperate during Müller glia‐mediated regeneration. To understand the molecular basis of microglial functions in Müller glia‐mediated regeneration, the transcriptional signature of retinal microglia was investigated during regeneration in zebrafish to identify unique “regeneration associated” transcripts (Mitchell et al. 2019).

We previously identified zebrafish cGMP phosphodiesterase‐6c (pde6c) mutants (Nishiwaki et al. 2008). Since Pde6c is a cone‐specific subunit of Pde6, which mediates phototransduction in cones, pde6c mutants show defects in photopic vision. In addition, pde6c mutants show progressive degeneration of cone photoreceptors by 6 months post‐fertilization (mpf), by which time most cones are eliminated (Figure 1B). Although Pde6c is specifically expressed in cones, rods also degenerate in pde6c mutants during embryonic and larval stages. However, rods recover progressively after Müller glia‐mediated regeneration starts at ~4 weeks post‐fertilization (wpf) and form a rod‐only photoreceptor cell layer in adult fish. We previously examined photoreceptor degeneration and Müller glia‐mediated cell proliferation in zebrafish aryl hydrocarbon receptor interacting protein‐like 1b (aipl1b) mutants (Iribarne et al. 2019), in which photoreceptor degeneration is mediated by Pde6c dysfunction (Iribarne et al. 2017). In aipl1b mutant retinas, photoreceptor apoptosis occurs at 2–3 wpf; however, Müller glia‐mediated regeneration does not start. After 5 wpf, photoreceptor apoptosis ceases, and Müller glia commence cell proliferation in aipl1b mutants. Thus, photoreceptor apoptosis and Müller glia‐mediated regeneration occur separately in aipl1b mutants. Accordingly, zebrafish mutants with pde6c dysfunction provide a unique platform for investigation of microglial states separately during photoreceptor degeneration and neuronal regeneration.

In this study, we conducted scRNA‐seq analysis of microglia in wild‐type sibling and pde6c mutant zebrafish retinas at 5 days post‐fertilization (dpf) and at 4 wpf, which correspond to onset of photoreceptor degeneration and commencement of Müller glia‐mediated regeneration, respectively. At 5 dpf, retinal microglia comprise three subclusters, representing either homeostatic, degeneration‐response, or stress‐response microglia. At 4 wpf, retinal microglia consist of four clusters, two of which appear predominantly in pde6c mutants and are localized to the photoreceptor layer. Furthermore, another cluster is prominently localized to the retinal stem cell niches and shares a similar transcriptome profile to NAM. Comparison of transcriptomic similarity between 4 wpf and 5 dpf microglial clusters revealed that each 4 wpf microglial cluster inherits homeostatic, degeneration‐response, and stress‐response characteristics of 5 dpf microglia in different combinations. Therefore, there is unique heterogeneity of microglia at the onset of Müller glia‐mediated regeneration. Taken together, our scRNA‐seq analysis reveals a dynamic change of retinal microglial states during the transition from photoreceptor degeneration to neuronal regeneration in zebrafish.

2. Methods

2.1. Fish Care and Ethics Statement

Zebrafish were maintained at 28°C on a 14:10 light: dark cycle, as previously described (Westerfield 2000). Embryos were reared in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.33 mM MgSO₄) until 4–6 dpf. To inhibit melanin biosynthesis for imaging and fluorescence‐activated cell sorting, phenylthiourea (PTU) was added to the E3 medium at a final concentration of 0.003%. All experiments conducted were approved by the Animal Care and Use Committee of Okinawa Institute of Science and Technology Graduate University (OIST), which is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC International). Approved protocol numbers: ACUP‐2023‐016, ACUP‐2023‐017, ACUP‐2023‐018, and ACUP‐2023‐019.

2.2. Fish Strains

Okinawa wild‐type (oki) was used as a standard wild‐type strain. roy orbison (roy) mutants (D'Agati et al. 2017) were introduced into other fish lines to prevent iridophore pigmentation. An allele of pde6c mutants, els ^rw76a , was used in this study (Nishiwaki et al. 2008). Tg[mpeg1.1:EGFP] ^oki053 and Tg[mfap4:tdTomato‐CAAX] ^oki083 were used to visualize microglia and macrophages (Ranawat and Masai 2021). Tg[ccl34b.1:eGFP] ^hkz035Tg was used to visualize microglia (Wu et al. 2018). Tg[mpeg1.1:NTR‐eYFP] ^w202 was used to deplete microglia upon treatment with MTZ (Petrie et al. 2014).

2.3. Screening Procedures to Identify pde6c Homozygous Mutant Fish

pde6c homozygous mutant fish were identified by genotyping. Genomic DNA was extracted from fins clipped from adult fish, or the tail fin and trunk from juveniles and larvae, respectively. A DNA fragment covering the pde6c mutation was amplified with PCR using primers indicated in Table S1. The amplified DNA fragment was digested with the restriction enzyme, FokI (Takara Bio), which cuts only the mutant‐derived genomic sequence (Nishiwaki et al. 2008). Alternatively, we used acridine orange (AO) for pre‐screening of zebrafish pde6c homozygous embryos, which show photoreceptor apoptosis after 4 dpf. For live imaging of AO‐labeled embryos, 4 dpf zebrafish roy mutant embryos were incubated at 28°C in E3 medium containing 4 μg/mL AO and 0.003% PTU for 1 h. They were then anesthetized in 0.02% tricaine and washed thoroughly in E3 medium at least thrice. After being kept undisturbed in fresh E3 medium with 0.003% PTU for 10 min, they were observed under an epifluorescence microscope (Leica). Third, the optokinetic response (OKR) was examined to detect blind larvae or juveniles in accordance with the previous report (Brockerhoff et al. 1995). As we reported previously (Nishiwaki et al. 2008), our pde6c homozygous mutant allele, pde6c ^rw76a, shows no visual response, including OKR, whereas their wild‐type siblings (pde6c heterozygous mutants and wild‐type) show normal OKR. Penetrance of OKR phenotype in pde6c ^rw76a mutation was almost 100%.

2.4. Quantitative Real‐Time PCR

RNA from fresh or snap‐frozen eyecups was extracted using Sepasol‐RNA I Super G (Nacalai Tesque). cDNA was synthesized using a ReverTra Ace qPCR RT Master Mix with gDNA Remover kit (Toyobo). Luna Universal qPCR Master Mix kit (New England Biolabs) was used to perform the qPCR reaction, with primers shown in Table S1. PCR reactions were run in an Applied Biosystems StepOnePlus Real‐Time PCR machine (ThermoFisher Scienetific). For qPCR of 5 dpf samples, 3 biological replicates, each of which pooled 20 eyes, were used. As technical replicates, each biological replicate or no‐template control was run in duplicate. For qPCR of 3–5 wpf samples, 4–6 biological replicates, each of which pooled 2–4 eyes, were used. As technical replicates, each biological replicate or no‐template control was run in duplicate for 3 wpf and in triplicate for 4 and 5 wpf. Data were analyzed using the 2^−ΔΔCt method (Livak and Schmittgen 2001), with the geometric mean of act2b and ef1a as a reference.

2.5. Microglia Depletion With the NTR‐MTZ System

To deplete microglia, we used the NTR/MTZ system. We treated Tg[mpeg1.1:NTR‐YFP; mfap4:tdTomato‐CAAX] or Tg[mfap4:tdTomato‐CAAX] transgenic embryos with E3 medium containing MTZ at 0, 5, 7.5, or 10 mM and 0.2% DMSO at 28°C in the dark from 3 to 5 dpf. These fish were genotyped for the pde6c locus as well as the presence of NTR‐YFP.

2.6. Application of the csf1r Inhibitor PLX3397

For PLX3397 treatment, 3.5 wpf Tg[ccl34b.1:eGFP] or Tg[mpeg1.1:EGFP] transgenic fish were maintained in a fish tank system with water containing PLX3397 at 0 or 500 nM and 0.05% DMSO, as described previously (Conedera et al. 2019), and maintained with daily water exchange until 5 wpf. To test depletion efficiency, Tg[ccl34b.1:eGFP] embryos were treated with E3 medium containing PLX3397 at 500 nM and 0.05% DMSO from 3 to 5 dpf.

2.7. Immunostaining and TdT‐Mediated dUTP Nick End Labeling (TUNEL)

Immunostaining of cryosections was performed as described previously (Imai et al. 2010). Primary and secondary antibodies were applied at an appropriate dilution in 10% goat serum (GS) or donkey serum, the latter of which was only used for labeling with goat anti‐Sox2 antibody and donkey anti‐goat secondary antibody in 0.1 M phosphate buffer (PB) containing 0.1% Triton X‐100 (0.1% PBTr). Information on antibodies and dilutions is provided in Table S1. Slides were washed in 0.1% PBTr. If necessary, nuclear staining was performed using Hoechst 33342 (Dojindo) at 1:1000, after which slides were rinsed in PB, and mounted in Fluoromount (Diagnostic BioSystems). For the TUNEL assay, In Situ Cell Death Detection Kit, TMR red or Fluorescein (Roche) was used according to the manufacturer's instructions. Confocal images were acquired with an LSM710 (Carl Zeiss), LSM780 (Carl Zeiss), or FV3000 (Evident).

To prepare retinal flat mounts at 3, 4, or 5 wpf, fish were euthanized in 0.04% tricaine. Eyes were dissected and transferred to L‐15 medium (ThermoFisher Scientific). Two sets of forceps were used to pull the eye apart to release the retina, which was then collected and fixed in 4% paraformaldehyde (PFA). Immunostaining was performed as described above, with minor changes. 0.1% Tween‐20 in 0.1 M phosphate buffer (0.1% PBT) was used to wash samples and dilute GS. Primary and secondary antibodies were diluted in 1% GS instead of 10% GS. After incubation with primary antibodies, samples were washed with 0.1% PBT and incubated with secondary antibodies for 3 h at room temperature. After washing, samples were mounted in RapiClear 1.49 (Sunjin Lab) and stored at 4°C. Confocal images were acquired with LSM780 (Carl Zeiss). Whole mount eyecups from 5 dpf fish were prepared in the same way as retinal flat mounts, except that 0.1% PBTr was used instead of 0.1% PBT, and secondary antibodies were applied overnight at 4°C. Confocal images were acquired with an LSM710 (Carl Zeiss), LSM780 (Carl Zeiss), or FV3000 (Evident).

2.8. Hybridization Chain Reaction Fluorescence In Situ Hybridization (HCR‐FISH)

Probe sets, buffers, and hairpins for HCR‐FISH were purchased from Molecular Instruments (Table S2) (Choi et al. 2010, 2018). Tg[mpeg1.1:EGFP] transgenic fish were used to prepare retinal sections with a cryostat (Cryostar NX70, ThermoFisher Scientific). HCR‐FISH was performed on retinal cryosections, according to the manufacturer's instructions. To distinguish between true signals and background autofluorescence, negative controls for HCR‐FISH were conducted with a mismatched hairpin1 that does not hybridize to either the target probe pairs or the desired hairpin2 (Table S3). mpeg1.1:EGFP signals were enhanced by labeling with anti‐GFP antibody. Nuclei were counter‐stained with Hoechst33342 (Dojindo) at 1:1000. Confocal images were acquired with an FV3000 (Evident). First, samples were imaged at 40× in VBF mode to acquire channels for mpeg1.1:EGFP, Hoechst, and the probe labeled with Alexa Fluor 647. Next, probes for 546 and 594 were imaged using the same objective and acquisition settings in lambda mode. These lambda mode images were spectrally deconvoluted using FluoView's blind unmixing or normal unmixing process, and then merged with the mpeg1.1:EGFP, Hoechst, and 647 image. If required, alignment of the two images was adjusted using the IMARIS Add Volume feature (Bitplane). Only mpeg1.1:EGFP‐positive cells that contained Hoechst‐labeled nuclei were used for HCR analysis, to define their locations precisely in the retina.

2.9. Live Imaging

For live imaging, embryos were anesthetized with tricaine and placed lateral side up on a grooved plate, in a droplet of E3 medium. They were mounted in 1.5% low‐melting agarose in E3 medium. After the low‐melting agarose solidified, more E3 medium was added on top and the eye was imaged with a water‐immersion objective lens on LSM710 (Carl Zeiss) or FV3000 (Evident).

2.10. Sphericity Analysis

IMARIS (v9.9.1, Bitplane) was used to render surface objects to represent microglia, based on (Nemes‐Baran and DeSilva 2021). Briefly, surfaces were generated based on absolute intensity of the mpeg1.1:EGFP channel and with a smoothing of 0.1 μm. Surfaces were manually unified or separated to match the fluorescence signal. Objects < 10 μm³ in volume were removed. Objects were also removed if a microglial cell was not associated with nuclear stain. Once surface generation was completed, objects were sampled based on location (associated with double cones or in the inner retina) and their sphericity measurements and object IDs were recorded.

2.11. Microglial Cluster Assignment

To classify microglia into a given cluster, surface objects based on the mpeg1.1:EGFP channel were created in IMARIS (v9.9.1, Bitplane), as described above in ‘Sphericity Analysis.’ Then, three spot types were generated for each of the 3 HCR‐FISH signals detected. To create a set of spots, the estimated XY diameter was set as 0.7 μm, and background subtraction was enabled. Object‐object statistics were enabled. Spots derived from signals away more than 0.1 μm from microglial surface objects were excluded by filtering. Furthermore, spots derived from signals corresponding to autofluorescence or noise were also manually excluded. The object ID for each microglia, its location in the retina, and the number of spots of each type within that object were recorded. Criteria for classification are provided in Table S4. Objects not meeting specified criteria for a given staining combination remained unclassified. 2–4 sections from 4 wild‐type sibling or pde6c mutant individuals each were examined. Information on signal counting and analysis are provided in Dataset 3 and Movie S1.

2.12. Tissue Homogenization and Fluorescence‐Activated Cell Sorting

Embryos were produced by pair‐wise crosses of Tg[mfap4:tdTomato‐CAAX] transgenic els ^+/− ; roy ^{+/− or −/−} fish and raised in E3 medium with 0.003% PTU. Embryos showing tdTomato‐CAAX‐positive microglia were sorted at 3 or 4 dpf. At 4 dpf, pde6c mutant embryos were separated from wild‐type sibling fish, using AO staining. At 4 wpf, we conducted the OKR assay to confirm that these separated pde6c homozygous mutants were blind. pde6c mutant and wild‐type sibling embryos were used for dissection of their eye cups at 5 dpf or 4 wpf. Dissected eye cups were homogenized to generate a single‐cell suspension as previously described (Mazzolini et al. 2018). Eye cups from 136 mutants and 140 wild‐type siblings were used at 5 dpf, as 5 dpf eye cups normally contain 40–50 microglia. Eye cups from 5 mutants and 5 wild‐type siblings (10 eyes each) were used at 4 wpf, as 4 wpf eye cups normally contain 400–500 microglia. Minor changes made to the protocol include precoating all 1.5‐mL tubes with Media A + 2% normal goat serum (NGS). Final resuspension of the cell pellet was done in 1× DPBS +2% NGS rather than in Media A + 2% NGS. Once resuspended, Sytox blue dead cell stain (Themo Fisher Scientific) was added at a dilution of 1:1000. The suspension was then strained through a 40‐μm cell strainer cap, into an ice‐cold FACS tube precoated with 1× DPBS +2% NGS. Information on these reagents is provided in Table S1.

Cell sorting was performed immediately on a BD FACSAria II or FACSAria III (BD Biosciences). Sorting parameters were standardized using roy fish bearing no fluorescence‐expressing transgene as a control, and mfap4:tdTomato‐CAAX transgenic fish as the test sample, at relevant ages (5 dpf or 4 wpf). Microglia/macrophages were sorted by gating on FSC/SSC scatter, Sytox blue dead stain, and tdTomato. Once settings to successfully sort live tdTomato⁺ cells were selected on both the FACSAria II and III, they were used unchanged to sort microglia from mutant and wild‐type sibling eyes for scRNA‐seq. Mutant and wild‐type sibling cells were sorted simultaneously on the FACSAria III and II, respectively, to minimize the time sorted microglia spent on ice before library preparation. Sorted cells were collected in tubes precoated in 1× DPBS +2% NGS. 2400–7100 live tdTomato⁺ cells were collected for each sample.

2.13. Single‐Cell RNA Sequencing

Isolated tdTomato⁺ microglia/macrophages were loaded onto Chromium Next GEM Chips (10× Genomics). Libraries for sequencing were then prepared using the Chromium Next GEM Single Cell 3′ Reagent Kits v3 or v3.1 (10× Genomics), following the manufacturer's protocols. Twelve cycles were used for cDNA amplification PCR and 14 cycles were used for indexing PCR. Sequencing was performed on an Illumina NovaSeq 6000. Raw fastq files were processed using the Cell Ranger pipeline from 10× Genomics, which performed alignment, filtering, barcode counting, and UMI counting, using the zebrafish reference genome GRCz11. Resulting filtered feature barcode matrices were used as input for downstream processing and visualization with Seurat v4 R package (Hao et al. 2021).

Each filtered feature barcode matrix (5d_sib, 5d_mut, 4w_sib, and 4w_mut) was converted into a Seurat object by CreateSeuratObject(). Quality filtering of each object was performed using parameters listed in Table S5. Datasets from 5d_sib and 5d_mut objects were combined by correcting technical batch effects while preserving true biological variation, using a similarity‐based method called Harmony, as were 4w_sib and 4w_mut objects. First, each dataset was normalized, scaled, and assayed for variable features (genes) with sctransform(), with variation from the percent of mitochondrial genes removed using regression. Second, Harmony (Korsunsky et al. 2019) was applied to integrate and batch‐correct each dataset, followed by principal component analysis (PCA). Third, clustering and non‐linear dimensional reduction with uniform manifold approximation and projection (UMAP) were performed using the FindNeighbors(), FindClusters(), and RunUMAP(). 5 dpf objects were clustered using 25 principal components (PCs) and a resolution of 0.5; however, 4 wpf objects were clustered using 25 PCs and a resolution of 0.6. FindAllMarkers() was used to identify DEGs for each cluster. DEGs between mutants and their wild‐type siblings in each cluster were identified using FindMarkers(), and significantly upregulated and downregulated genes (adjusted p value < 0.05) were subjected to GO (biological process) enrichment analysis using the clusterProfiler package for R (Yu et al. 2012).

2.14. Cluster Mapping and Id Prediction

To compare transcriptomic similarities between 5d_Mg and 4w_Mg clusters, we utilized Seurat's single‐cell reference mapping and cell type classification algorithm (Stuart et al. 2019). FindTransferAnchors() using 5d_Mg as reference and 4w_Mg as query was run to find a set of anchor genes between reference and query objects in the first 30 PCs. Then TransferData() was run to classify 4w_Mg based on 5d_Mg cell type annotations. Predicted cell types were plotted on UMAP using RunUMAP(). Fractions of each predicted cell type in 4w_Mg clusters were visualized using heatmap.2() from the gplots package. Additionally, Mphage and pMg/pMphage clusters were excluded from the 5 dpf and 4 wpf scRNA‐seq datasets, and reference mapping and cell type classification were rerun with only Mg clusters using the same parameters. Prediction scores of each 4w_Mg cell returned from TransferData() were visualized with ternary diagrams using the Ternary package.

Comparisons between microglia/macrophage populations in pde6c mutant and wild‐type sibling eyes at 4 wpf (4w_Mg) and microglia populations identified in the mid−/hindbrain of juvenile zebrafish at 28 dpf (JMs), including synaptic‐region‐associated microglia (SAM) and neurogenic‐associated microglia (NAM) (Silva et al. 2021), were performed using the same workflow. The count matrix and metadata of juvenile microglial cells from (Silva et al. 2021) were downloaded from GEO: GSE164772 and Seurat objects were constructed using CreateSeuratObject(). Both 4w_Mg and JM datasets were individually normalized and scaled with NormalizeData() and ScaleData() with default parameters. Then PCA dimensionality reduction was run by RunPCA(). Next, FindTransferAnchors() and TransferData() using the first 30 PCs of JMs as a reference and 4w_Mg as query classified 4w Mgs based on the JMs cell type annotations (JM1‐4). Predicted cell types were plotted on UMAP using RunUMAP(). Fractions of each predicted cell type in 4w_Mg clusters were visualized with heatmap.2() from the gplots package.

2.15. Quantification and Statistical Analysis

Statistical analysis for qPCR and histological experiments was performed in GraphPad Prism 9.5.1. RNA sequencing data was analyzed in R (ver. 4.4.1). Statistical details of experiments (including sample sizes, dispersion measures, statistical tests used) are described in figure legends.

2.16. AI Disclosure

No generative AI or AI‐assisted technologies were used in the writing, research, or content creation of this manuscript.

3. Results

3.1. Microglia Respond to Photoreceptor Degeneration and Accumulate in the ONL in Zebrafish pde6c Mutants at 5 Dpf

We examined zebrafish pde6c mutants to determine how microglia respond to photoreceptor degeneration during embryonic stages. We previously generated zebrafish transgenic lines, Tg[mpeg1.1:EGFP] and Tg[mfap4:tdTomato‐CAAX], both of which visualize microglial precursors and mature microglia (Ranawat and Masai 2021). Here, we focused on one pde6c mutant allele, pde6c ^rw76a, which carries a missense mutation, Met175Arg, in the cone‐specific α’ subunit of Pde6 (Nishiwaki et al. 2008). We combined pde6c mutants with the Tg[mpeg1.1:EGFP] transgenic line and examined retinal apoptosis by TUNEL (Figure 1C,D). At 3 dpf, apoptosis was rare in the ONL of both wild‐type siblings and pde6c mutants. At 4 dpf, apoptosis was drastically increased in the ONL of pde6c mutants. A higher level of apoptosis was also detected in pde6c mutants at 5 dpf, although the difference was not significant. Thus, photoreceptors start to undergo apoptosis at 4 dpf in pde6c mutants. Second, we examined the microglial response (Figure 1C). At 3 dpf, microglia were mostly located in the RGCL and INL, but rare in the ONL, in both wild‐type siblings and pde6c mutants. At 4 dpf, a small number of microglia migrated into the ONL in pde6c mutants. By 5 dpf, many microglia migrated into the ONL, including the OS, in pde6c mutants. The number of microglia in the ONL was higher in pde6c mutants than in wild‐type siblings at both 4 and 5 dpf (Figure 1E). Third, we examined the number of microglia in wild‐type sibling and pde6c mutant retinas at 4 and 5 dpf. Live imaging of Tg[mpeg1.1:EGFP] transgenic retinas revealed that the total number of ocular microglia was higher in pde6c mutants than in wild‐type siblings at 5 dpf (Figure S1A,B). This was confirmed by whole‐mount immunofluorescence (Figure 1F). However, there was no significant difference in ocular microglia number between pde6c mutants and wild‐type siblings at 4 dpf (Figure S1A,B). Taken together, these data suggest that retinal microglia respond to photoreceptor degeneration at 4 dpf, increase in number, and migrate into the ONL by 5 dpf in pde6c mutants.

3.2. Microglia Eliminate Dying Photoreceptors by Phagocytosis at 5 Dpf

To examine microglial response to dying photoreceptors, we labeled pde6c mutant Tg[mfap4:tdTomato‐CAAX] transgenic retinas with acridine orange, which visualizes apoptotic cells, including engulfed dead cells (Abrams et al. 1993). We confirmed that microglia phagocytose dying photoreceptors in pde6c mutants at 5 dpf (Figure 1G). Next, we applied the NTR/MTZ system (Petrie et al. 2014) to deplete microglia. First, we applied MTZ at three concentrations (5, 7.5, and 10 mM) to double transgenic Tg[mpeg1.1:NTR‐YFP; mfap4:tdTomato‐CAAX] embryos from 3 to 5 dpf, and examined the number of microglia in their retinas (Figure S1C). Microglia were effectively eliminated by 7.5 mM and 10 mM MTZ (Figure S1D,E). Next, we applied 10 mM MTZ to Tg[mfap4:tdTomato‐CAAX] pde6c mutant and wild‐type sibling embryos with or without the transgene Tg[mpeg1.1:NTR‐YFP], and examined apoptosis by TUNEL (Figure S1F). In the presence of MTZ, retinal microglia number was higher in non‐NTR transgenic pde6c mutants than in non‐NTR transgenic wild‐type siblings. However, this increased number was reduced in Tg[mpeg1.1:NTR‐YFP] transgenic pde6c mutants to an equivalent level of non‐NTR transgenic wild‐type siblings (Figure S1G). Consistently, in the presence of MTZ, the number of TUNEL⁺ cells in the ONL was further increased in Tg[mpeg1.1:NTR‐YFP] transgenic pde6c mutants, compared with non‐NTR transgenic pde6c mutants (Figure S1H). Thus, apoptotic photoreceptors failed to be eliminated in microglia‐depleted pde6c mutants, indicating that microglia remove degenerating photoreceptors at 5 dpf.

Next, we examined expression of microglia signature genes in wild‐type sibling and pde6c mutant eyes at 5 dpf. apoeb and apoc1 genes encode apolipoproteins, which regulate lipid and cholesterol trafficking (Babin et al. 1997). The p2ry12 gene encodes a metabolic purinergic receptor that functions as an apoptotic signal sensor (Blume et al. 2020; Haynes et al. 2006), and is a marker of homeostatic microglia (Butovsky and Weiner 2018). C1q is a complement component and a marker of homeostatic microglia (Keren‐Shaul et al. 2017). interleukin‐1β (il1b) encodes IL‐1β, which shows pro‐inflammatory effects in neurodegenerative diseases (Voet et al. 2019). Secreted phosphoprotein 1 (Spp1) is the most highly upregulated gene in DAM (Lan et al. 2024). Accordingly, il1b and spp1 are pro‐inflammatory genes. Compared with wild‐type siblings, pde6c mutants showed significant upregulation of microglial signature genes, apoeb and apoc1, as well as homeostatic genes, c1qb and p2ry12, whereas expression of pro‐inflammatory genes, il1b and spp1, was unchanged (Figure 1H). Thus, microglia are active in phagocytosis of dying photoreceptors and seem to maintain a homeostatic state in pde6c mutants at 5 dpf.

3.3. Activated Microglia Accumulate in the ONL at 4 Wpf When Müller Glia Commence Photoreceptor Regeneration

Proliferation of Müller glia was observed at 5 wpf in zebrafish pde6c mutants (Nishiwaki et al. 2008), but not at 3 wpf in zebrafish aipl1b mutants (Iribarne et al. 2019), which lack Pde6c activity (Iribarne et al. 2017). Therefore, it is likely that Müller glia‐mediated neuronal regeneration starts between 3 and 5 wpf in pde6c mutants (Figure 2A). First, we examined mRNA expression of seven regulators of retinal regeneration, ascl1a, hbegfa, insm1a, mmp9, pax6a, pcna, and sox2, in wild‐type sibling and pde6c mutant retinas at 3, 4, and 5 wpf, using qPCR analysis (Figure 2B). As in wild‐type siblings, at 3 wpf, mRNA expression of these seven genes had not increased in pde6c mutant retinas. At 4 wpf, their expression was upregulated in pde6c mutant retinas, with especially significant upregulation of insm1a, pax6a, sox2, and pcna. At 5 wpf, significant upregulation was observed for ascl1a and pcna, but expression levels of hbegfa, insm1a, mmp9, pax6a, and sox2 had dropped back to sibling levels. Thus, Müller glia‐mediated regeneration is initiated at 4 wpf and proceeds at 5 wpf. Consistently, we labeled 4 wpf wild‐type sibling and pde6c mutant retinas with anti‐PCNA and zrf‐1 antibodies, which visualize proliferative cells and Müller glia, respectively (Figure 2C). zrf‐1 signal‐associated PCNA+ cells increased in number and formed a cluster in the INL, suggesting that Müller glia enter cell proliferation in pde6c mutants at 4 wpf. To evaluate Müller glia‐mediated proliferation, we also labeled wild‐type sibling and pde6c mutant retinas with anti‐Sox2 and anti‐PCNA antibodies at 5 wpf (Figure 2D). Sox2 regulates Müller glial reprogramming and proliferation in the regenerating zebrafish retina (Gorsuch et al. 2017). Sox2 is weakly expressed in Müller cells as well as a subset of amacrine cells in undamaged zebrafish retina; however, Sox2 expression is markedly increased in proliferating Müller glia after retinal damage (Gorsuch et al. 2017). Consistently, the number of Sox2‐ and PCNA‐double positive cells in the INL and ONL, which correspond to proliferative Müller glia and their‐derived neural progenitor cells, respectively, was increased in pde6c mutant retinas, compared with wild‐type sibling retinas (Figure 2E), although the difference in INL cell number was not significant.

FIGURE 2.

Microglia upregulate expression of homeostatic and inflammatory genes when Müller glia commence regeneration. (A) Timeline of the onset of Müller glia‐mediated neuronal regeneration in pde6c mutants. (B) qPCR of retinal regeneration genes in the eye cup at 3, 4, and 5 wpf. Statistical significance was evaluated by unpaired t‐test: means ± SD. ns, p > 0.05; *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. (C) Four wpf wild‐type sibling and pde6c mutant retinas labeled with anti‐PCNA (green) and zrf‐1 (magenta) antibodies. Dotted lines indicate the OPL. Bottom panels are higher magnification images of rectangles shown in the upper panels. Yellow arrows indicate zrf‐1 signals associated with PCNA‐positive cells, indicating proliferative Müller glia. Scale bars: 20 μm. (D) Five wpf wild‐type sibling and pde6c mutant retinas labeled with anti‐PCNA (green) and anti‐Sox2 (red) antibodies. Nuclei are counterstained with Hoechst (blue). Bottom panels are higher magnification images of rectangles shown in upper panels. Yellow arrows and arrowheads indicate Sox2 and PCNA‐double positive cells in the INL and ONL, respectively. ONL, outer nuclear layer; INL, inner nuclear layer; RGCL, retinal ganglion cell layer. Scale bars: 30 μm. (E) Graph of the number of Sox2⁺ Müller glia in the ONL, INL, and both nuclear layers (total) per retinal section of wild‐type siblings and pde6c mutants. Each dot indicates the average number of 2–3 sections from one fish. Two‐way ANOVA with Sidak multiple comparison test: means ± SD, ns, p > 0.05; ****p ≤ 0.0001. (F) Flat‐mount retinas of wild‐type siblings and pde6c mutants at 4 wpf. Microglia are visualized with Tg[mpeg1.1:EGFP] (green). Double cone‐type photoreceptors are labeled with zpr1 antibody (magenta). Top panels indicate confocal z‐axis images of wild‐type sibling and pde6c mutant retinas. Three optical slice positions corresponding to (1) the subretinal area (rod OS and RPE), (2) the zpr1⁺ cone layer, and (3) the inner nuclear layer (INL) are indicated by arrows. Bottom panels show three optical slices. Many microglia are observed only in the subretinal area and the cone layer in pde6c mutant retinas (yellow asterisks). Scale bar: 70 μm. (G) Microglial density is calculated as the number of microglia per mm². Each dot indicates an individual fish. Nested t‐tests (two tailed): means ±SD, *p ≤ 0.05, ***p ≤ 0.001; ****p ≤ 0.0001. (H) Magnified images of 4 wpf wild‐type sibling and pde6c mutant Tg[mpeg1.1:EGFP] transgenic retinas labeled with zpr1 antibody. Top and bottom panels indicate the cone layer and INL of flat‐mounted retinas, respectively. Two higher magnification images indicating microglial cells phagocytosing zpr1⁺ cone material in cone layers of pde6c mutants are presented. Scale bars: 20 μm. (I) Sphericity of microglia in the inner retinal layer and cone layer of wild‐type siblings and pde6c mutants. Sphericity of each microglial cell was calculated using surface‐rendered objects prepared with IMARIS software. Each point represents one object. Brown‐Forsythe and Welch ANOVA with Dunnett's T3 multiple comparisons test: Box‐and‐whisker plot, ns, p > 0.05; *p ≤ 0.05, ****p ≤ 0.0001. (J) PCR of microglia‐related genes in wild‐type sibling and pde6c mutant eyes at 3, 4, and 5 wpf. Four‐six biological replicates of 2–4 eyes each were tested. ND: not detected. Unpaired t‐tests: means ± SD. ns, p > 0.05; *p ≤ 0.05, ***p ≤ 0.001; ****p ≤ 0.0001.

Next, we examined the distribution of microglia in wild‐type sibling and pde6c mutant retinas at 4 wpf. To globally survey microglial distribution in whole retina, we used flat‐mount Tg[mpeg1.1:EGFP] transgenic retinas labeled with zpr1 antibody, which visualizes double cone‐type photoreceptors (Larison and Bremiller 1990), and evaluated the density of microglia in three retinal areas: (1) subretinal area corresponding to OS and RPE, (2) zpr1⁺ cone cell layer, and (3) the INL (Figure 2F). In wild‐type siblings, microglia were mostly located in the INL, and were seldom in the cone layer and subretinal area. On the other hand, in pde6c mutants, round‐shaped microglia markedly accumulated in the cone layer and subretinal area, whereas microglia were mildly increased in the INL. The density of microglia in the whole retina was significantly higher in pde6c mutants than in wild‐type siblings at 3, 4, and 5 wpf (Figure 2G). Furthermore, in pde6c mutants, microglia in the cone layer showed more amoeboid shapes, in contrast to their ramified shapes in inner retina, and often phagocytosed zpr1‐positive cones (Figure 2H). Consistently, microglia in the cone layer exhibited greater sphericity than those of the inner retina, which is consistent with the morphology of activated microglia (Nayak et al. 2014; Stence et al. 2001) (Figure 2I). Thus, many microglia continue to reside in the cone layer and phagocytose dying cones in pde6c mutants at 4 wpf.

To examine whether pro‐inflammatory or homeostasis‐linked genes are expressed in microglia during the neuronal regeneration stage, we examined expression of microglial signature genes, apoeb, apoc1, c1qb, p2ry12, il1b, and spp1 in wild‐type siblings and pde6c mutant retinas at 3, 4, and 5 wpf (Figure 2J). Expression of p2ry12 was significantly upregulated in pde6c mutant retinas at all stages, which is consistent with phagocytic microglia consuming cones in pde6c mutants (Figure 2H). pde6c mutants also showed upregulation of apoeb and apoc1 after 4 wpf, indicating activation of lipid metabolism. Notably, expression of the proinflammatory cytokine, il1b, was upregulated in pde6c mutants after 4 wpf, consistent with a previous report (Lu and Hyde 2024). Upregulation of the p2ry12 gene in pde6c mutants may be consistent with a previous report that anti‐inflammatory microglia are more numerous than pro‐inflammatory microglia in retinas of zebrafish aipl1b mutants (Iribarne and Hyde 2022).

3.4. scRNA‐Seq Analysis Identifies Three Non‐Proliferative Microglial Subclusters at 5 Dpf

To understand the role of microglia in photoreceptor degeneration and regeneration, we applied scRNA‐seq to investigate microglial states in wild‐type sibling and pde6c mutant retinas at 5 dpf and 4 wpf, which correspond to the onset of photoreceptor degeneration and the start of Müller glia‐mediated regeneration, respectively. We combined pde6c mutant or wild‐type sibling fish with the Tg[mfap4:tdTomato‐CAAX] transgene. Then, we dissected eye cups from 5 dpf larval and 4 wpf juvenile fish, collected tdTomato⁺ cells using flow cytometry, and conducted scRNA‐seq using the 10× Chromium platform (Figure 3A).

FIGURE 3.

Identification of microglial clusters at the onset of photoreceptor degeneration. (A) Workflow of scRNA‐seq experiment using zebrafish retinas at 5 dpf and 4 wpf. (B) UMAP plot showing three microglial clusters, two macrophage clusters, and one proliferative microglia and macrophage‐mixed cluster at 5 dpf. Clustering resolution = 0.5. (C) Expression of microglia markers (apoeb, apoc1, p2ry12, ccl34b.1), macrophage markers (c1qb, lygl1, marco), and proliferative markers (mki67, cdk1, ccna2) per cluster at 5 dpf. (D) Fraction of each microglia and macrophage cluster in wild‐type siblings and pde6c mutants. (E) Heatmap of the top 10 enriched genes for each microglial cluster, 5d_Mg1/2/3, at 5 dpf. Wild‐type sibling and pde6c mutant cells are split for each subcluster. Degeneration‐response markers (ckbb, apoc1, apoeb, lgasl9l), stress‐response markers (ubb, junbb, hsp70.2, hsp70.3), and homeostatic markers (csf1ra, srgap2, elmo1, mcamb) are highlighted in red, green, and blue colors, respectively. (F) Dotplot analysis of genes related to oxidative phosphorylation (atp5mc1, cox7c, ndufa4l, mt‐co1), ribosomal biogenesis (rps13, rpl17), proinflammatory cytokine (il1b), homeostatic genes (csf1ra and csf1rb), and galectin genes (lgals2a, lgals9l1, and lgals9l3), which represent neuroprotective microglia, high protein synthesis, proinflammatory microglia, homeostatic microglia, and repair of degenerative issues, respectively. Microglial groups highly expressing each gene category are marked by color‐coded squares.

In the scRNA‐seq analysis on mfap4:tdTomato⁺ cells for wild‐type sibling (n = 666) and pde6c mutant (n = 508) at 5 dpf, we identified 6 clusters (Figures 3B and S2A). Using specific markers of microglia (apoeb, apoc1, p2ry12, ccl34b.1) (Mazzolini et al. 2020; Wu et al. 2020; Zhou et al. 2023) and macrophages (c1qb, lygl1, marco) (Zhou et al. 2023), we identified three microglia clusters (5d_Mg1, 5d_Mg2, 5d_Mg3), two macrophage clusters (5d_Mphage1, 5d_Mphage2), and one proliferating microglia‐ and macrophage‐mixed cluster (5d_pMg/Mphage) (Figure 3C). 5d_pMg/Mphage was prominently found in pde6c mutants (Figures 3D and S2B), which is consistent with the increase of retinal microglia from 4 to 5 dpf (Figure S1B). Furthermore, 5d_Mg1 was overrepresented in pde6c mutants, compared to wild‐type siblings. On the other hand, 5d_Mg2 and 5d_Mg3 were underrepresented (Figures 3D and S2B).

Next, we focused on three non‐proliferative microglial clusters and examined their expression profiles. 5d_Mg1 microglia were enriched in microglia signature genes, apoeb and apoc1, as well as neuroprotective markers, ckbb and lgals9l1 (Figure 3E, Dataset 1). Ckbb is highly expressed in microglia associated with axon tracts with remyelination in mice (Masuda et al. 2019). Zebrafish Lgals9l1 is a homologue of mouse Lgals9 (Galectin‐9), which is expressed in microglia and macrophages and promotes oligodendrocyte remyelination and functional recovery after stroke in mice (Han et al. 2024). Since 5d_Mg1 fraction increased in pde6c mutants compared with wild‐type siblings (Figure 3D), 5d_Mg1 is likely to be a cluster of degeneration‐response microglia, but expresses neuroprotection‐associated genes. On the other hand, 5d_Mg3 microglia are enriched for csf1ra, elmo1, mcamb, and srgap2 (Figure 3E). In zebrafish, csf1ra is a major Csf1r in the embryonic stage (Oosterhof et al. 2018; Wu et al. 2018), and Csf1r is a homeostatic marker of mouse microglia (Keren‐Shaul et al. 2017). Elmo1 promotes engulfment of apoptotic cells by microglia in zebrafish (van Ham et al. 2012). Mcamb facilitates cell migration and protrusion in zebrafish (Ye et al. 2013). Srgap2 is highly expressed in microglia in zebrafish and promotes their ramification (Uribe‐Salazar et al. 2024). Thus, 5d_Mg3 is a cluster of homeostatic, ramified microglia. Lastly, 5d_Mg2 microglia share expression of the top 10 enriched genes of 5d_Mg3, although their expression level is lower than that of 5d_Mg3 (Figure 3E). In contrast to 5d_Mg3, 5d_Mg2 microglia highly express immediate early response genes such as ubb, junbb, hsp70.2, and hsp70.3 (Figure 3E), suggesting that 5d_Mg2 comprises stress‐response microglia.

To precisely characterize each cluster, we examined expression of genes belonging to 10 functional categories: (1) oxidative phosphorylation (Brischigliaro and Zeviani 2021), (2) ribosomal biogenesis (Gay et al. 2022), (3) lipid metabolism (apoeb, apoc1, insig1, and lipg), (4) lysosome regulation (rxraa, bhlhe40/41, and cathepsins), (5) Jak/Stat pathway, (6) pro‐ and anti‐inflammatory cytokines, (7) stress response, (8) homeostatic microglial genes (csf1r, c1q, srgap2, p2ry12), (9) Notch signaling, and (10) others (hbegfb, mpp1, mrc1b, lgals2a, lgals9l1/l3) (Figure S2C). We also prepared dotplots of selected representative genes of each category (Figure 3F). Both oxidative phosphorylation‐ and ribosomal biogenesis‐related genes were highly expressed in 5d_Mg1, especially in pde6c mutants, but not active in 5d_Mg2 and 5d_Mg3, especially in wild‐type siblings (Figure 3F). Indeed, GO enrichment analysis confirmed that oxidative phosphorylation is upregulated in 5d_Mg1 of pde6c mutants (Figure S2D). This is consistent with our proposal that 5d_Mg1 comprises anti‐inflammatory microglia because a metabolic shift toward active oxidative phosphorylation is associated with anti‐inflammatory microglia (Devanney et al. 2020). Indeed, the link between the upregulation of oxidative phosphorylation‐related genes and anti‐inflammatory roles of immune cells was also suggested in macrophage‐mediated regeneration of zebrafish lateral line (Denans et al. 2022). Consistently, expression of inflammatory cytokine, il1b, was not detected in 5d_Mg1 and low in homeostatic microglia, 5d_Mg3 (Figure 3F). On the other hand, il1b expression was highly elevated in the stress‐response microglia, 5d_Mg2, indicating that 5d_Mg2 comprises pro‐inflammatory microglia (Figure 3F). Homeostatic microglial markers, csf1ra and csf1rb, were highly expressed in 5d_Mg3 (Figure 3F). Interestingly, only csf1rb expression was reduced for pde6c mutants. Both csf1r genes were silenced in degeneration‐response microglia, 5d_Mg1, consistent with previous reports that csf1r expression is downregulated in DAM in mice (Butovsky and Weiner 2018; Keren‐Shaul et al. 2017). Interestingly, Galectin proteins, lgals2a and lgals9l1/l3, were upregulated only in pde6c mutant microglia, especially for 5d_Mg1 (Figure 3F). lgals2a (previously known as drgal1‐l2 (Vasta et al. 2004)) mRNA expression is initiated in proliferating Müller glia in adult zebrafish retinas after light‐mediated injury, and that lgals2a mRNA is also expressed in microglia (Craig et al. 2010). Knockdown of lgals2a does not affect entry of Müller glia into cell proliferation and cone regeneration, but reduced rod regeneration, suggesting its role in rod regeneration (Craig et al. 2010). Lgals9 suppresses NLRP3‐inflammasomes, leading to reduction of IL‐1β release from macrophages (Liu and Stowell 2023), and promotes oligodendrocyte remyelination and functional recovery after stroke in mice (Han et al. 2024). Taken together, this GOI heatmap shows that each microglial subcluster has a unique pattern of metabolism.

3.5. All Three Microglial Clusters Move Toward the Photoreceptor Layer in pde6c Mutants at 5 Dpf

Next, we investigated where three microglial clusters, 5d_Mg1, 5d_Mg2, and 5d_Mg3, are located in zebrafish retina using HCR‐FISH (Figure S3A) (Choi et al. 2010, 2018). To identify microglial clusters, we visualized expression of apoeb, rxraa, and snx33 mRNA in retinal sections of Tg[mpeg1.1:EGFP] transgenic embryos (Figure S3B). apoeb serves as a pan‐microglial marker. rxraa labels 5d_Mg2 and 5d_Mg3, and snx33 is enriched in 5d_Mg3 (Figures 3E and 4A,B). Microglia were also labeled with anti‐GFP antibody. Next, to precisely classify microglia into these three clusters, surface objects of mpeg1.1:EGFP‐positive microglia were created from 3D scanning images using the surface rendering tool of IMARIS software (Bitplane) (Movie S1). Since fluorescent HCR signals were usually detected as spotted signals, spot objects representing HCR signals of apoeb, rxraa, and snx33 mRNAs in microglial surface objects were manually counted (Figure S3B, Movie S1). Then, each microglial cell was classified into one of three clusters (5d_Mg1, 5d_Mg2, and 5d_Mg3) based on the combination of apoeb, rxraa, and snx33 mRNA expression (Figure S3C). Finally, we determined the retinal areas (CMZ, RGCL, IPL, INL, OPL, ONL and OS/RPE) in which each microglial cluster is preferentially localized (Figure 4C–E).

FIGURE 4.

Localization of each 5d_Mg cluster in the retina. (A) Feature plot showing expression of apoeb, rxraa, and snx33 genes at 5 dpf. (B) Classification strategy of 5 dpf microglial clusters, based on HCR‐FISH with apoeb, rxraa, and snx33 RNA probes. (C) Color‐code assignment to seven retinal regions at 5 dpf. (D) HCR‐FISH of 5 dpf microglial clusters, 5d_Mg1/2/3, in Tg[mpeg1.1:EGFP] transgenic wild‐type sibling and pde6c mutant retinas with apoeb, rxraa, and snx33 RNA probes. Higher magnification images show microglia located in the CMZ/RGCL/IPL/INL area for wild‐type siblings and the INL/OPL/ONL/OS area for pde6c mutants. Dotted lines demarcate the interface between each retinal area. Yellow line indicates the outline of microglia defined by Tg[mpeg1.1:EGFP] expression. Scale bars: 10 μm. (E) Bar charts showing the distribution of each cluster across retinal areas in wild‐type siblings and pde6c mutants, using color‐code assignment defined in (C). Microglia were sampled from four fish for each group using 3–4 sections per fish. (F) Graphical summary of scRNA‐seq analysis of zebrafish 5 dpf retinal microglia. At 5 dpf, three microglial clusters, 5d_Mg1, 5d_Mg2, and 5d_Mg3, are identified as degeneration‐response microglia, stress‐response microglia, and homeostatic microglia, respectively, in accordance with their transcriptomic profiles. In wild‐type sibling retinas, 5d_Mg1 is broadly localized in the CMZ, RGCL, INL, and ONL, whereas 5d_Mg2 and 5d_Mg3 are preferentially localized in the CMZ and RGCL. In pde6c mutant retinas, all three Mg clusters move toward the OPL, ONL, OS/RPE, and upregulate lgals2a/lgasl9l1. In addition, 5d_Mg1 in pde6c mutants upregulate oxidative phosphorylation (OXPHOS)‐related genes, suggesting a neuroprotective role.

5d_Mg1 microglia in wild‐type siblings were distributed across all retinal areas except the ONL and OS/RPE. However, in pde6c mutants, the number of microglia localized in the CMZ and RGCL was decreased, whereas the number of microglia localized in the ONL and OS/RPE was drastically increased (Figure 4E). 5d_Mg2 microglia were localized in the CMZ or RGCL in wild‐type siblings. However, in pde6c mutants, the number of microglia in the CMZ and RGCL was drastically reduced, whereas most microglia accumulated in either the INL, OPL, ONL, or OS/RPE (Figure 4E). 5d_Mg3 microglia were predominantly localized in the CMZ in wild‐type siblings. However, in pde6c mutants, 5d_Mg3 microglia localized in the CMZ were drastically reduced, but those localized in the INL, OPL, ONL, and OS/RPE were markedly increased (Figure 4E). Thus, all three microglial clusters move toward degenerating photoreceptors in pde6c mutants (Figure 4F).

3.6. scRNA‐Seq Analysis Identifies Four Non‐Proliferative Microglial Subclusters at 4 Wpf

To characterize the role of microglia in Müller glia‐mediated regeneration in pde6c mutants, we performed scRNA‐seq analysis on ocular microglia of 4 wpf pde6c mutant and wild‐type sibling fish carrying the Tg[mfap4:tdTomato‐CAAX] transgene. We collected tdTomato⁺ cells (sibling: n = 973; mutant: n = 366) and identified 8 clusters (Figures 5A and S4A). Using specific markers of microglia and macrophages, we identified four microglial clusters (4w_Mg1, 4w_Mg2, 4w_Mg3, 4w_Mg4), three macrophage clusters (4w_Mphage1, 4w_Mphage2, 4w_Mphage3), and one proliferating microglia‐ and macrophage‐mixed cluster (4w_pMg/Mphage) (Figure 5B). The 4w_pMg/Mphage cluster makes up a similar fraction of microglia in pde6c mutants and their wild‐type siblings (Figures 5C and S4B), suggesting that cell proliferation is not activated in microglia of pde6c mutants. Compared with wild‐type siblings, 4w_Mg1 was decreased in pde6c mutants. However, 4w_Mg2 did not change its fraction size in pde6c mutants. On the other hand, 4w_Mg3 and 4w_Mg4 were expanded in pde6c mutants (Figures 5C and S4B).

FIGURE 5.

Identification of microglial clusters at the start of neuronal regeneration. (A) UMAP plot showing four microglial clusters, three macrophage clusters, and one proliferative microglia and macrophage‐mixed cluster at 4 wpf. Clustering resolution = 0.6. (B) Expression of microglia markers (apoeb, apoc1, p2ry12, ccl34b.1), macrophage markers (c1qb, lygl1, marco), and proliferative markers (mki67, cdk1, ccna2) per cluster at 4 wpf. (C) Fraction of each microglia and macrophage cluster in wild‐type siblings and pde6c mutants. (D) Heatmap of the top 10 enriched genes of each microglial cluster at 4 wpf. Degeneration‐response markers (apoeb, apoc1, glula, g0s2, fabp11a, ribosomal proteins), stress‐response marker (rsrp), and homeostatic markers (elmo1, srgap2, rbpjb, rxraa) are indicated. Microglial groups highly expressing 4w_Mg3‐enriched genes are indicated with light blue squares. Microglial groups highly expressing 4w_Mg4‐enriched genes are indicated with red squares. (E) Dotplot analysis of genes related to oxidative phosphorylation, ribosomal biogenesis, lipid metabolism, lysosome regulation, Jak/Stat, il1b, stress response, homeostatic genes, Notch signaling, and galectin genes. Microglial groups highly expressing each category of genes are marked with color‐coded squares.

Next, we focused on four non‐proliferative microglial clusters and examined their expression profiles. 4w_Mg1‐enriched genes were rbpjb, srgap2, elmo1, rxraa, and rsrp1 (Figure 5D). In addition, 4w_Mg1 highly expressed csf1ra, mcamb, hipk2, cblb, ubb, junbb, ier2a, and ier2b (Dataset 2). These genes are highly enriched in 5d_Mg3 and 5d_Mg2 (Figure 3E), suggesting that 4w_Mg1 shares characteristics of homeostatic microglia 5d_Mg3 and stress‐response microglia 5d_Mg2 (Figure 5D). 4w_Mg2‐enriched genes were apoeb, apoc1, glula, g0s2, lgals9l1 (Figure 5D, Dataset 2), all of which were enriched in 5d_Mg1, the degeneration‐response cluster (Figure 3E). 4w_Mg3 highly expressed ckbb, g0s2, fabp11a, lgals9l1, rps27.2, slc25a5, and rpl35 (Figure 5D, Dataset 2), all of which are top 10 enriched in 5d_Mg1 (Figure 3E). Lastly, 4w_Mg4 highly expressed ribosomal biogenesis genes, but also expressed apoc1, slc25a5, ckbb, and apoeb (Figure 5D, Dataset 2), which are likewise enriched in 5d_Mg1 (Figure 3E). Thus, 4w_Mg2, 4w_Mg3, and 4w_Mg4 differentially inherit characteristics of degeneration‐response microglia 5d_Mg1. To confirm this, we mapped 4w_Mg cells onto 5d_Mg clusters by comparing transcriptomic similarity (Figure 6A). 4w_Mg1 shares a transcriptomic profile with both 5d_Mg2 and 5d_Mg3. 4w_Mg2 is similar to both 5d_Mg1 and 5d_Mg3. 4w_Mg3 resembles both 5d_Mg1 and 5d_Mg2. 4w_Mg4 shares a transcriptomic profile predominantly with 5d_Mg1 (Figure 6A). Thus, 4w_Mg2 and 4w_Mg3 also share characteristics of homeostatic 5d_Mg3 and stress‐response 5d_Mg2, respectively. Indeed, mapping of individual 4w_Mg cells onto a triangular diagram of 5d_Mg clusters revealed that each 4w_Mg cluster consists of a more heterogeneous population, even though each 4w_Mg cluster is biased toward a certain group of 5d_Mg cluster characteristics, as mentioned above (Figure 6B).

FIGURE 6.

Clustering, and mapping of 4w_Mg clusters on 5d_Mg clusters. (A) Mapping of 4w_Mg clusters onto 5d_Mg clusters. Left panel shows the original UMAP of 4wpf_Mg clusters. Middle panel shows the “predicted.id” UMAP of 4w_Mg clusters, in which each 4w_Mg cell is assigned to either 5d_Mg cluster, whose transcriptomic profile shows the highest similarity to that of the targeted 4w_Mg cell. Right panel shows the breakdown of percentage of 4w_Mg cells assigned to each 5d_Mg cluster. (B) Plotting of individual 4w_Mg1/2/3/4 cells on the triangular diagram of the three 5d_Mg clusters, using transcriptomic similarity percentages of individual 4w_Mg1/2/3/4 cells to each 5d_Mg cluster, calculated with the cluster mapping analysis shown in (A). The left panel shows merged plotting of all 4w_Mg cells. Right panels show plotting of each 4w_Mg cluster, in which plotting density is also shown. 4w_Mg1 cells (red circles) are mapped around the edge that connects vertices of 5d_Mg3 and 5d_Mg2. 4w_Mg2 cells (green circles) are mapped around the edge that connects vertices of 5d_Mg1 and 5d_Mg3. 4w_Mg3 cells (blue circles) are mapped around the edge that connects vertices of 5d_Mg1 and 5d_Mg2. Almost all 4w_Mg4 cells (purple circles) are mapped to the vertex of 5d_Mg1. These spatial plotting data are consistent with cluster mapping data shown in (A); however, importantly, each 4w_Mg cell population shows heterogeneity on inheritance rate of characteristics of each 5d_Mg cluster. (C) Visualization of il1b expression levels in each 4w_Mg cell mapped onto a triangular diagram of 5d_Mg1/2/3 clusters. Expression level is indicated with red (high)‐green (low) color code. 4w_Mg cells highly expressing il1b are located around the vertex showing the highest transcriptomic similarity to stress‐response microglia 5d_Mg2. Thus, il1b expression level is determined by transcriptomic similarity to stress‐response microglia even in 4w_Mg cells. (D) Summary of the transcriptomic similarity between 5d_Mg clusters and 4w_Mg clusters. 4w_Mg1 inherits characteristics of homeostatic 5d_Mg3 and stress‐response 5d_Mg2. 4w_Mg2 inherits characteristics of degeneration‐response 5d_Mg1 and homeostatic 5d_Mg3. 4w_Mg3 inherits characteristics of degeneration‐response 5d_Mg1 and stress‐response 5d_Mg2. 4w_Mg4 purely inherits characteristics of degeneration‐response 5d_Mg1.

To precisely determine characteristics of each cluster, we examined expression of the same 10 categories of genes used for 5 dpf microglial analysis (Figure S4C). We also prepared a dotplot of selected representative genes from each category (Figure 5E). First, oxidative phosphorylation‐ and ribosomal biogenesis‐related genes were downregulated in 4w_Mg1 (Figure 5E), which is similar to those of homeostatic 5d_Mg3 and stress‐response 5d_Mg2 (Figure 3F). However, in contrast to degeneration‐response 5d_Mg1, which highly expresses both categories of genes, oxidative phosphorylation genes were moderately activated in 4w_Mg2, but suppressed in 4w_Mg3; however, ribosomal biogenesis genes were significantly suppressed in 4w_Mg2, but enhanced in 4w_Mg3 (Figure 5E). In addition, both categories of genes were highly expressed in 4w_Mg4 (Figure 5E). Such heterogeneous expression patterns in 4w_Mg2/3/4 are explained by cluster mapping data showing that 4w_Mg2, 4w_Mg3 and 4w_Mg4 have characteristics of degeneration‐response/homeostatic merged, degeneration‐response/stress‐response merged, and purely degeneration‐response microglia, respectively (Figure 6A,B). Second, lipid metabolism genes, apoeb and apoc1 had low expression in 4w_Mg1, but were highly upregulated in both 4w_Mg2 and 4w_Mg4 (Figure 5E). Interestingly, only apoc1 expression was highly upregulated in 4w_Mg3 (Figure 5D). Furthermore, apoeb expression was downregulated, but apoc1 expression was maintained at a high level in the pde6c mutant population of 4w_Mg4 (Figure 5D,E). insig1 is an ER membrane‐localized cholesterol sensor protein (Kacher et al. 2022). lipg is an endothelial lipase, and promotes lipid biosynthesis (J. E. Yu et al. 2018). Both insig1 and lipg genes were highly expressed in 4w_Mg3 (Figures 5E and S4C). However, expression of both genes was moderately low in 4w_Mg1, and markedly suppressed in 4w_Mg2 and 4w_Mg4. On the other hand, both were upregulated in the pde6c mutant population of 4w_Mg1 and 4w_Mg2 (Figures 5E and S4C), which was also confirmed by UMAP and DEG analysis (Figure S4D,E). Thus, expression of these lipid metabolism genes is dynamically changed between the wild‐type and pde6c mutant environments. Third, lysosome‐related genes such as cathepsins were highly enriched in 4w_Mg2, compared with other 4 wpf microglial clusters (Figure 5E). Since cathepsin family genes are moderately expressed in all the three 5 dpf microglial clusters (Figure S2C), some special role may be assigned to 4w_Mg2. Fourth, the five categories, (1) Jak/Stat signaling, (2) pro‐ and anti‐inflammatory cytokines, (3) stress‐response genes, (4) homeostatic microglial markers, and (5) Notch signaling are downregulated in degeneration‐response 5d_Mg1, but moderately or highly upregulated in homeostatic 5d_Mg3 and stress‐response 5d_Mg2 (Figure S2C). However, such a clear difference was not observed between 4w_Mg1, 4w_Mg2 and 4w_Mg3 (Figure 5E and S4C). Therefore, heterogeneity of expression of these categories of genes is reduced between 4w_Mg1/2/3. This supports data that 4w_Mg1 and 4w_Mg2 share characteristics of homeostatic microglia 5d_Mg3, and that 4w_Mg1 and 4w_Mg3 share characteristics of stress‐response microglia 5d_Mg2 (Figures 5D and 6A,B). However, as in 5d_Mg1, expression of these categories of genes was markedly suppressed in 4w_Mg4 (Figure 5E), which is also consistent with data that 4w_Mg4 purely inherits the transcriptomic profile of 5d_Mg1 (Figure 6A,B).

Importantly, the pro‐inflammatory cytokine, il1b, was highly upregulated in 4w_Mg1 and 4w_Mg3, and moderately expressed in 4w_Mg2; however, it was markedly downregulated in 4w_Mg4 (Figure 5E). Therefore, we conclude that 4w_Mg1/3 are pro‐inflammatory microglia. Since oxidative phosphorylation genes are highly enriched in 4w_Mg4 (Figure 5E), it is possible that 4w_Mg4 is categorized as anti‐inflammatory microglia. Next, we visualized il1b expression level of each 4w_Mg cell on the triangular diagram of 5d_Mg clusters because il1b was highly expressed in stress‐response 5d_Mg2 (Figure 3F). Interestingly, il1b expression in 4w_Mg cells closely correlates with their transcriptomic similarity to stress‐response 5d_Mg2 (Figure 6C). Thus, it is very likely that stress‐response signature genes determine il1b expression levels even in 4wpf microglia. Next, regeneration‐associated galectin gene, lgals2a, was suppressed in the wild‐type population of all 4 wpf microglial subtypes, but upregulated in the pde6c mutant population, especially in 4w_Mg4 (Figure 5E). lgals9l1 was highly expressed in both wild‐type and mutant populations of 4w_Mg3, but its expression was upregulated in the pde6c mutant population of the other three clusters (Figure 5E). Upregulation of lgals2a and lgals9l1 in pde6c mutants suggests that remodeling of the photoreceptor cell layer environment may be activated by 4 wpf microglia, especially 4w_Mg4, facilitating Müller glia‐mediated regeneration. Taken together, these data indicate dynamic rearrangement of microglial states from 5 dpf to 4wpf (Figure 6D).

3.7. 4w_Mg3 and 4w_Mg4 are Specifically Generated in pde6c Mutants

Next, we investigated localization of four microglial clusters, 4w_Mg1, 4w_Mg2, 4w_Mg3 and 4w_Mg4, in the zebrafish retina, using HCR‐FISH (Choi et al. 2010; Choi et al. 2018). We visualized expression of four genes, apoeb, rxraa, bzw2 and lipg. apoeb serves as a pan‐microglial marker. rxraa labels three clusters, 4w_Mg1–Mg3. bzw2 and lipg mark 4w_Mg2 and 4w_Mg3, respectively (Figure 7A,B). Three combinations of these mRNA probes, namely (1) apoeb, rxraa, and bzw2/lipg mixture, (2) apoeb, rxraa, and bzw2, (3) apoeb, rxraa, and lipg, were applied to retinal sections of Tg[mpeg1.1:EGFP] transgenic wild‐type sibling and pde6c mutant embryos (Figure S3D). Next, HCR‐FISH and signal counting within microglial‐surface‐rendered objects were performed as described earlier. In accordance with the combination of apoeb, rxraa, bzw2 and lipg mRNA expression (Figure S3E), we determined in which retinal areas (CMZ, CMZ‐adjacent, RGCL, IPL, INL, OPL, ONL and OS/RPE) each microglial cluster is preferentially localized (Figure 7C–E).

FIGURE 7.

Localization of each 4w_Mg cluster in the retina. (A) Feature plot showing expression of apoeb, rxraa, bzw2, and lipg genes at 4 wpf. (B) Classification strategy of 4 wpf microglial clusters, based on HCR‐FISH with apoeb, rxraa, bzw2, and lipg RNA probes. (C) Color‐coded assignment to eight retinal regions at 4 wpf. (D) HCR‐FISH of 4 wpf microglial clusters, 4wpf_Mg1/2/3/4, in Tg[mpeg1.1:EGFP] transgenic wild‐type and pde6c mutant retinas with apoeb, rxraa, bzw2, and lipg RNA probes. Higher magnification images show microglia located in IPL/INL/OPL area for 4w_Mg1, CMZ/RGC/IPL/INL area for 4w_Mg2, and INL/OPL/ONL/OS area for 4w_Mg3 and 4w_Mg4. Dotted lines demarcate the interface between retinal areas. Yellow line indicates the outline of microglia defined by Tg[mpeg1.1:EGFP] expression. Scale bars: 10 μm. (E) Bar charts showing the distribution of each cluster across retinal areas in wild‐type siblings and pde6c mutants, using color‐code assignment defined in (C). Microglia were sampled from four fish for each group using 3–4 sections per fish. (F) Graphical summary of scRNA‐seq analysis of zebrafish 4 wpf retinal microglia. At 4 wpf, four microglial clusters are identified as homeostatic/stress‐response microglia (4w_Mg1), degeneration‐response/homeostatic microglia (4w_Mg2), degeneration‐response/stress‐response microglia (4w_Mg3), and degeneration‐response microglia (4w_Mg4) in accordance with their transcriptomic profile. Both 4w_Mg3 and 4w_Mg4 are rare in wild‐type siblings, but appear and approach the photoreceptor layer in pde6c mutants, indicating pde6c mutant‐specific microglia clusters. Furthermore, 4w_Mg3 serve as pro‐inflammatory microglia. lipg is enriched in 4w_Mg3; however, all other Mg subtypes in pde6c mutants upregulate lipg expression. 4w_Mg4 upregulates lgals2a in pde6c mutants, suggesting putative anti‐inflammatory microglia. 4w_Mg2 approaches the photoreceptor layer in pde6c mutants.

4w_Mg1 microglia were localized in the RGCL, INL, and OS/RPE in wild‐type siblings, whereas 4w_Mg1 microglia localized in the RGCL were markedly decreased in pde6c mutants (Figure 7D,E). In both wild‐type siblings and pde6c mutants, 4w_Mg1 microglia in the INL were mostly ramified (Figure 7D). Thus, 4w_Mg1 microglia are normally in a homeostatic condition. In wild‐type siblings, more than 50% of 4w_Mg2 microglia were localized in either the CMZ or the CMZ‐adjacent area, which is located at the interface between the CMZ and the central differentiated retinal region (Figure 7C). Together, we refer to the CMZ and CMZ‐adjacent areas as CMZa. The remaining 4w_Mg2 microglia were localized in the INL and ONL. In pde6c mutants, 4w_Mg2 microglia continued to be localized in the CMZa, INL and ONL; however, 4w_Mg2 microglia localized in the OPL and OS/RPE were increased (Figure 7D,E), suggesting that some 4w_Mg2 microglia move toward the OS area. 4w_Mg3 and 4w_Mg4 were rarely observed in wild‐type siblings; however, they emerged abundantly in pde6c mutants and mostly localized in the OS/RPE (Figure 7D,E). We noted that many 4w_Mg1 and 4w_Mg2 cells upregulate lipg expression in pde6c mutants (Figure S4D,E). Specific expression of bzw2 in 4w_Mg2 enables us to distinguish 4w_Mg3 from 4w_Mg2. However, in pde6c mutants, it is challenging to distinguish 4w_Mg3 from 4w_Mg1 that shows higher levels of lipg expression, because 4w_Mg1 cells are also found in the OS region. It is likely that lipg‐positive 4w_Mg1 and 4w_Mg2 are consecutive state microglia moving toward 4w_Mg3 in pde6c mutants, because their spatial distances were close in UMAP data (Figure 7A). Spatial distribution of 4w_Mg clusters is therefore dynamically changed between wild‐type sibling and pde6c mutant retinas (Figure 7F).

3.8. 4w_Mg2 and 4w_Mphage1 Exhibit Transcription Profiles Similar to NAM and SAM, Respectively

Our HCR‐FISH data indicate that more than 50% of 4w_Mg2 are located in the CMZa (Figure 7D,E). Cluster mapping data show that 4w_Mg2 shares characteristics of degeneration‐response 5d_Mg1 and homeostatic 5d_Mg3 (Figure 6A,B,D). 5d_Mg3 is mostly localized in the CMZ (Figure 4D,E). Thus, 4w_Mg2 may be related to neurogenesis. In zebrafish, microglia associated with neurogenic regions in the optic tectum and midbrain were identified as NAM (Silva et al. 2021). This NAM cluster highly expresses lysosome‐related genes, including cathepsin family proteins (Silva et al. 2021). Indeed, 4w_Mg2 also expresses cathepsin genes (Figure 5E and S4C), so we examined the similarity between 4w_Mg2 and NAM. We mapped our 4wpf scRNA‐seq data onto a published scRNA‐seq dataset of mpeg1:EGFP⁺ cells in the zebrafish mid‐ and hindbrain, which contain a JM1 cluster identified as NAM (Silva et al. 2021) (Figure S5A and S5B). Comparison of transcriptomic similarity revealed that 41.4% of 4w_Mg2 and 22.3% of 4w_Mg1 are classified into the JM1 cluster. Thus, it is likely that 4w_Mg2 is more closely related to NAM. Furthermore, 23 genes of the top 50 enriched in the JM1 were found in the top 50 enriched genes of 4w_Mg2, whereas no gene was overlapped in the top 50 enriched genes between JM1 and 4w_Mg1 (Figure S5C). These data suggest that 4w_Mg2 is NAM in zebrafish retina.

Silva et al. also identified synaptic region‐associated microglia as SAM, which corresponds to JM4 (Silva et al. 2021). Surprisingly, 4w_Mphage clusters of our dataset were assigned to JM4; especially 4w_Mphage1 was the highest matching in transcriptomic similarity (94.6%) (Figure S5A,B). We compared the top 50 enriched genes between 4w_Mphage1/2/3 and JM4. The top 50 enriched genes of JM4 overlapped with those of 4w_Mphage1 at 26.0% (n = 13/50), 4w_Mphage2 at 4.0% (n = 2/50), and 4w_Mphage3 at 30.0% (n = 15/50) (Figure S5D). However, almost all overlapping genes of 4w_Mphage3 were ribosomal proteins (Figure S5D). Nine of 13 overlapping non‐ribosomal genes were in the top 10 of 4w_Mphage1 (Figure S5D). Therefore, it is likely that 4w_Mphage1 shares characteristics of SAM. SAM contains blood vessel‐associated macrophages in zebrafish mid‐ and hindbrain (Silva et al. 2021). We previously reported that microglial precursors, which express macrophage markers such as lygl1, enter the optic cup through ocular blood vessels, probably hyaloid artery, at the early stage of zebrafish development (Ranawat and Masai 2021). Recently, it was also reported that in zebrafish, microglial precursors migrate into the optic cup through the basal surface of retinal CMZ along the superficial ocular vasculature, especially the ventral radial vessel (VRV), in the later stage of development after choroid fissure closure (Zhan et al. 2025). We observed that mpeg1.1:EGFP⁺ cells expressing lygl1 mRNA, but not apoeb mRNA, were positioned along the basal border of the CMZ in both wild‐type sibling and pde6c mutant retinas at 4 wpf (Figure S6). Since lygl1 is also a marker of microglial/macrophage precursors, it is possible that 4w_Mphage1 contains microglial/macrophage precursors migrating toward the retinal CMZ along ocular blood vessels, which may share a SAM‐like transcriptomic profile.

3.9. Csf1r Inhibitor Does Not Prevent Müller Glial From Proliferating, but Suppresses Their Migration Toward the ONL in pde6c Mutants at 4 Wpf

In adult zebrafish retinas with damage, either an inflammation suppressor drug, dexamethasone, or genetic ablation of microglia inhibits Müller glia‐mediated regeneration (Iribarne and Hyde 2022; Silva et al. 2020; Zhang et al. 2020). However, csf1r inhibitor, PLX3397, which affects survival of microglia, does not inhibit Müller glial proliferation in adult zebrafish retinas with laser injury induction, although later steps of neuronal regeneration were affected (Conedera et al. 2019). We are interested in this PLX3397‐induced effect, which is different from the effects of dexamethasone and genetic ablation of microglia. We used PLX3397 to deplete microglia in pde6c mutants from 3.5 wpf onward, and examined Müller glial proliferation at 5 wpf (Figure 8A). Indeed, PLX3397 treatment did not inhibit Müller glial proliferation, but allowed Müller glia‐derived progenitors to form a large cluster in the INL in pde6c mutants, compared with DMSO‐treated pde6c mutants (Figure 8B), consistent with a previous report (Conedera et al. 2019).

FIGURE 8.

Distribution and classification of retinal microglia in DMSO or PLX3397‐treated wild‐type siblings and pde6c mutants. (A) Timeline of PLX3397 treatment in juvenile fish. (B) Five wpf retinas of wild‐type siblings treated with DMSO and pde6c mutants treated with DMSO or PLX3397, labeled with anti‐PCNA (green) and zrf‐1 (magenta) antibodies. Dotted lines indicate the OPL. Bottom panels show higher magnification images of proliferating Müller glia indicated by dotted rectangles in the top panel. In DMSO‐treated wild‐type siblings, proliferative Müller glia are sparse. In DMSO‐treated pde6c mutants, most proliferative Müller glia migrate to the ONL (yellow arrowheads). In PLX3397‐treated pde6c mutants, proliferative Müller glia are increased, but mostly located in the INL (yellow arrows). Scale bars: 20 μm. (C) Five wpf retinas of wild‐type sibling treated with DMSO and pde6c mutant treated with DMSO or PLX3397, labeled with anti‐PCNA (green) and anti‐Sox2 (red) antibodies. Nuclei are counterstained with Hoechst (blue). Bottom panels are higher magnification images of the rectangles shown in upper panels. Yellow arrows and arrowheads indicate Sox2 and PCNA‐double positive cells in INL and ONL, respectively. Scale bars: 30 μm. (D) Graph of the number of PCNA⁺; Sox2⁺ Müller glia in ONL, INL and both layers (total) per retinal section of DMSO‐treated wild‐type siblings and DMSO or PLX3397‐treated pde6c mutants. Each dot represents an average of 2–3 sections from one fish. Two‐way ANOVA with Sidak multiple comparison test: means ± SD, ns, p > 0.05; *p ≤ 0.05, **p ≤ 0.005; ***p ≤ 0.001. (E) Graph of the fraction of ONL‐resident PCNA⁺; Sox2⁺ Müller glia in the total PCNA⁺; Sox2⁺ Müller glia population. Two‐way ANOVA with Tukey multiple comparison test: means ± SD, ns, p > 0.05; *p ≤ 0.05. (F) A possible model of PLX3397‐mediated suppression of regeneration of Müller glia in pde6c mutants. PLX3397 may inhibit the proliferation rate of Müller glia or their migration toward the ONL, leading to the accumulation of proliferative Sox2⁺ Müller glia in the INL of pde6c mutants. (G) Retinal sections of 5 wpf Tg[mpeg1.1:EGFP] transgenic wild‐type zebrafish treated with DMSO or PLX3397. Dotted lines mark the retinal area. Arrowheads indicate a representative apoeb⁺  microglial cell, whose apoeb, rxraa, and lipg/bzw2 expression is visualized with HCR‐FISH and shown in magnified panels on the right. Scale bar: 50 μm for left large panels, 10 μm for magnified images. (H) Bar chart showing the distribution of apoeb ⁺ microglia across retinal layers in wild‐type siblings. 4w_Mg cells localized in RGCL, INL, and OS/RPE are decreased in PLX3397 treated wild‐type siblings. (I) Bar chart showing classification of apoeb ⁺ microglia into ‘4w_Mg1/CMZa‐resident 4w_Mg2/4w_Mg4’ or ‘Other’ clusters in wild‐type siblings. The 4w_Mg1 fraction is reduced by PLX3397 treatment. (J) Retinal sections of 5 wpf Tg[mpeg1.1:EGFP] transgenic pde6c mutant zebrafish treated with DMSO or PLX3397, as described in (G). (K) Bar chart showing the distribution of apoeb ⁺ microglia across retinal layers in pde6c mutants. 4w_Mg cells localized in OS/RPE are decreased in PLX3397 treated pde6c mutants. (L) Bar chart showing classification of apoeb ⁺ microglia into 4w_Mg1, CMZa‐resident 4w_Mg2, 4w_Mg4 or ‘Other’ cluster in pde6c mutants. The 4w_Mg1 fraction is maintained in the presence of PLX3397. On the other hand, the ‘Other’ fraction is reduced by PLX3397 treatment.

To precisely evaluate this PLX3397‐induced effect, we labeled DMSO‐ and PLX3397‐treated pde6c mutant retinas with anti‐Sox2 and anti‐PCNA antibodies at 5 wpf (Figure 8C). We rarely detected Sox2‐ and PCNA‐double positive cells in DMSO‐treated wild‐type sibling retinas, except for small numbers of Sox2‐ and PCNA‐double positive cells localized in the dorsal ONL (Figure 8C). On the other hand, the number of Sox2⁺ Müller glia and their‐derived neural progenitor cells increased in DMSO‐treated pde6c mutant retinas, especially in the ONL (Figure 8C,D). In PLX3397‐treated pde6c mutant retinas, the number of Sox2⁺ cells was higher in DMSO‐treated wild‐type sibling retinas, but lower in DMSO‐treated pde6c mutant retinas (Figure 8C,D). Thus, Müller glial cell proliferation is not fully activated in PLX3397‐treated pde6c mutant retinas. Interestingly, the number of Sox2⁺ INL cells was mildly higher in PLX3397‐treated pde6c mutants than in DMSO‐treated pde6c mutants, whereas the number of Sox2⁺ ONL cells was significantly lower in PLX3397‐treated pde6c mutants than in DMSO‐treated pde6c mutants (Figure 8D). Consistently, the fraction of Sox2⁺ ONL cells in the total Sox2⁺ cells was significantly lower in PLX3397‐treated pde6c mutants than either in DMSO‐treated wild‐type sibling or DMSO‐treated pde6c mutants, suggesting that migration of proliferating Müller glia or their‐derived neural progenitors toward the ONL is suppressed by PLX3397‐treatment (Figure 8C,E). Thus, although PLX3397 treatment does not inhibit entry of Müller glia into a proliferative state, it may suppress their proliferation rate or migration of their progenitors to the ONL (Figure 8F).

3.10. PLX3397 Preferentially Eliminate Microglial Cells Highly Expressing csf1r

Since csf1r is a marker of homeostatic microglia, it is likely that csf1ra/csf1rb are expressed differentially in each microglia subtype. It is possible that PLX3397 selectively eliminates microglial subtypes with high susceptibility to csf1r functions. To examine this possibility, we conduct HCR‐FISH of 5 wpf Tg[mpeg1.1:EGFP] transgenic wild‐type retinas treated with either DMSO or PLX3397 with apoeb, rxaa, lipg and bzw2 RNA probes (Figure 8G). Compared with DMSO‐treated wild‐type retinas, PLX3397‐treated wild‐type retinas showed a severe reduction of RGCL‐ and OS/RPE‐resident microglia and a mild reduction in INL‐resident microglia, but rather maintained microglia in CMZa (Figure 8H). Furthermore, the classification strategy of 4w_Mg clusters (Figure 7B) revealed that 4w_Mg1 is markedly reduced in PLX3397‐treated wild‐type retinas, compared with DMSO‐treated wild‐type retinas (Figure 8I). This is consistent with our previous observation that 4w_Mg1 is preferentially localized in RGCL, INL and OS/RPE of wild‐type sibling retinas (Figure 7E). Furthermore, PLX3397 treatment at 500 nM eliminated retinal microglia by 5 wpf (Figure S7A,B), consistent with a previous report (Conedera et al. 2019). However, microglial cells remaining in the retina after PLX3397 treatment are predominantly located in or adjacent to the CMZ (Figure S7A,C). Since more than 50% of 4w_Mg2 are localized in the CMZ in wild‐type sibling retinas (Figure 7D), microglia remaining after PLX3397 treatment are likely to be 4w_Mg2, which we confirmed (Figure 8G). Since 4w_Mg1 shows a higher level of csf1ra/csf1rb expression compared with other 4w_Mg clusters including 4w_Mg2 (Figure 5E), these data suggest that PLX3397 treatment preferentially eliminates microglial cells that highly express csf1ra/csf1rb. We also examined the effect of PLX3397 induced in embryonic microglia at 5 dpf. PLX3397 treatment at 500 nM reduced the number of retinal microglia at 5 dpf; however, this difference was not significant (Figure S7D,E). 5d_Mg1 shows a very low level of csf1ra/csf1rb expression, whereas 5d_Mg2 and 5d_Mg3 show intermediate and high levels of csf1ra/csf1rb expression, respectively (Figure 3F). It is possible that 5d_Mg1 and also 5d_Mg2 are less sensitive to PLX3397 treatment. The survival of 5d_Mg1/2 may mask PLX3397‐mediated elimination of 5d_Mg3.

3.11. PLX3397 Treatment Does Not Eliminate 4w_Mg1, but Reduces OS/RPE‐Localized Microglia in pde6c Mutants

To examine how PLX3397 treatment modifies microglial state and distribution in pde6c mutant retinas at the regeneration stage, we conducted HCR‐FISH of 5 wpf DMSO‐ or PLX3397‐treated Tg[mpeg1.1:EGFP] transgenic pde6c mutant retinas with apoeb, rxaa, lipg, and bzw2 RNA probes (Figure 8J). Compared with DMSO‐treated pde6c mutant retinas, PLX3397‐treated pde6c mutant retinas showed a reduction of microglia localized in the OS/RPE (Figure 8K). Surprisingly, the classification strategy of 4w_Mg clusters (Figure 7B) revealed that compared with DMSO‐treated pde6c mutant retinas, the fraction of 4w_Mg1 was not decreased in PLX3397‐treated pde6c mutant retinas (Figure 8L). Since csf1rb expression of 4w_Mg1 is lower in pde6c mutants than in wild‐type siblings (Figure 5E), 4w_Mg1 may become less sensitive to PLX3397 in pde6c mutants. Our classification strategy of 4w_Mg clusters combined with HCR‐FISH using all four RNA probes (apoeb, rxaa, lipg, and bzw2) makes it difficult to distinguish between 4w_Mg2 and 4w_Mg3. Therefore, we placed them in the category called “Other” (Figure 8I,L), except the CMZa‐localized 4w_Mg2. Compared with DMSO‐treated pde6c mutant retinas, the “Other” fraction was decreased in PLX3397‐treated pde6c mutants, suggesting that either central retina‐resident 4w_Mg2 or 4w_Mg3 is reduced in pde6c mutants.

4. Discussion

In zebrafish retinas that suffer damage or neuronal degeneration, Müller glial cells are reprogrammed to generate neural progenitor cells, which subsequently regenerate retinal neurons (Lahne et al. 2020). Microglia‐induced inflammation is required for Müller glia‐mediated neuronal regeneration in zebrafish (Nagashima and Hitchcock 2021). Therefore, it is important to understand how microglial states are regulated during Müller glia‐mediated neuronal regeneration in zebrafish. Zebrafish pde6c mutants provide an excellent model that enables us to investigate microglial states in photoreceptor degeneration and Müller glia‐mediated neuronal regeneration separately. Here, we conducted scRNA‐seq analysis on ocular microglia in wild‐type and pde6c mutant retinas at 5 dpf and 4 wpf and revealed the configuration of microglia states at each stage.

At 5 dpf, non‐proliferative microglia consist of three clusters: degeneration‐response 5d_Mg1, stress‐response 5d_Mg2, and homeostatic 5d_Mg3. Previous scRNA‐seq analysis of adult mouse retinas revealed that retinal microglia normally form one large cluster but are split into six subtypes, one of which specifically approaches degenerating photoreceptors, suggesting a special microglia subtype that responds to photoreceptor degeneration in mice (O'Koren et al. 2019). In pde6c mutants, the fraction of 5d_Mg1 is increased, whereas 5d_Mg2 and 5d_Mg3 are reduced. Thus, 5d_Mg1 is likely to be a major cluster that phagocytoses degenerating photoreceptors. However, our HCR‐FISH analysis revealed that more than 70% of all three microglial clusters in pde6c mutants accumulate in either the OPL, ONL, or OS area. Therefore, we did not find a specific microglial cluster associated with dying photoreceptors. Instead, all three clusters respond to dying photoreceptors at 5 dpf. At present, we do not know the reason for the difference in microglial response to degenerating photoreceptors between zebrafish and mice. It is possible that each cluster of zebrafish 5 dpf embryonic microglia is less specialized in contrast to microglia in adult mice. Indeed, among 4 wpf zebrafish microglia, two pde6c mutant‐specific microglial clusters, namely 4w_Mg3 and 4w_Mg4, seem to be more specialized to approach photoreceptor layer niches.

GOI expression analysis revealed that all 5d_Mg clusters show specific transcriptome patterns. For example, genes of oxidative phosphorylation and ribosomal biogenesis are highly expressed in 5d_Mg1, but not in 5d_Mg3, indicating that mitochondrial respiration and protein translation are active in degeneration‐response microglia, but silent in homeostatic microglia. Expression of homeostatic markers, csf1ra/csf1rb and srgap2, is high in 5d_Mg3, but very low in 5d_Mg1, which is consistent with our observation that ramified and amoeboid shapes are prominent in 5d_Mg3 and 5d_Mg1, respectively. One surprising finding is that the global pattern of GOI expression is very similar between 5d_Mg2 and 5d_Mg3, indicating that stress‐response microglia may be derived from homeostatic microglia. Interestingly, pro‐inflammatory cytokine, il1b, is highly expressed only in stress‐response 5d_Mg2. Such heterogeneity of transcriptome patterns between microglial clusters suggests some task‐allocation mechanism between microglial clusters at the onset of photoreceptor degeneration. We still do not know whether all three microglial clusters moving toward dying photoreceptors maintain such task specialization, because there are no gross differences of expression pattern between wild‐type and mutant populations of each 5d_Mg cluster. An exception is found in galectin family proteins, lgals2a and lgals9l1/l3. In wild type, expression of these genes is very low in all three 5d_Mg clusters. However, both lgals2a and lgals9l1/l3 genes are highly upregulated in the mutant population of 5d_Mg clusters. Thus, these galectins are specifically upregulated in mutants, which may serve some common function in tissue remodeling to prevent inflammation after elimination of dying photoreceptors in pde6c mutants.

At 4 wpf, non‐proliferative microglia consist of four clusters: homeostatic/stress‐response merged 4w_Mg1, degeneration‐response/homeostatic merged 4w_Mg2, degeneration‐response/stress‐response merged 4w_Mg3, and purely degeneration‐response 4w_Mg4. HCR‐FISH analysis revealed that 4w_Mg3 and 4w_Mg4 clusters specifically appear in pde6c mutants and are mostly localized in the OS area, suggesting that they are pde6c mutant photoreceptor‐associated microglia. So, wild‐type retinas mainly accommodate two clusters, 4w_Mg1 and 4w_Mg2. 4w_Mg1 is localized in the RGCL, INL, and OS area, whereas more than 50% of 4w_Mg2 is localized in the CMZ‐adjacent area. Furthermore, 4w_Mg2 has characteristics of NAM, highly expressing apoeb and apoc1 and cathepsin family genes. However, GOI expression analysis revealed that global transcriptome patterns of other category genes (oxidative phosphorylation, ribosomal biogenesis, Jak/Stat, cytokines, stress response, homeostatic markers, and Notch signaling) are relatively similar between 4w_Mg1 and 4w_Mg2, suggesting less heterogeneity between 4w_Mg1 and 4w_Mg2. This is supported by cluster mapping data showing that 4w_Mg1 and 4w_Mg2 share the transcriptomic profile of homeostatic 5d_Mg3. Furthermore, il1b expression levels of 4w_Mg cells closely correlate with their transcriptomic similarity of stress‐response 5d_Mg2, raising the possibility that stress‐response signature genes predominantly determine the expression level of il1b. One interesting question is why clustering of 4 wpf microglial cells does not match the transcriptomic similarity to three 5d_Mg axes, namely degeneration‐response, stress‐response, and homeostasis. To address this question, it is necessary to investigate the role of the top 10 enriched genes in each of 4w_Mg clusters.

In this study, we identified two pde6c mutant‐specific microglial clusters, 4w_Mg3 and 4w_Mg4. In 4w_Mg3, ribosomal biogenesis is activated, although oxidative phosphorylation genes are downregulated. 4w_Mg3 also highly expresses (1) lipid metabolism regulators, apoc1, insig1, and lipg, (2) pro‐inflammatory cytokine, il1b, (3) stress‐response genes, and (4) galectin genes, lgals9l1/l3. In 4w_Mg3, lysosome‐related genes such as cathepsins and homeostatic genes csf1ra and srgap2 are downregulated. From this profiling, 4w_Mg3 is likely to be pro‐inflammatory. Interestingly, expression of the top 10 enriched genes of 4w_Mg3, rgcc, fabp11a, lipg, cidec, and eef1 family genes, is low in wild‐type microglia of other 4w_Mg clusters, but highly upregulated in mutant populations of all other 4w_Mg clusters. These data suggest that all microglial clusters respond to the 4 wpf pde6c mutant environment and activate expression of 4w_Mg3 enriched genes. Thus, these 4w_Mg3 signature genes are strong candidates to promote Müller glia‐mediated neuronal regeneration. On the other hand, 4w_Mg4 has different characteristics. In 4 wpf_Mg4, oxidative phosphorylation genes and apoeb are highly upregulated, whereas other category genes (lysosome functions, Jak/Stat, cytokines, stress‐response genes, homeostatic genes, notch signaling) are markedly downregulated. Regeneration‐associated galectin, lgals2a, is upregulated in the mutant population of 4w_Mg4, so it is likely that 4w_Mg4 serves some anti‐inflammatory function; however, further investigation will be necessary to clarify its role in neuronal regeneration.

Interestingly, 4w_Mg2 has a similar transcriptome profile to NAM in zebrafish mid‐ and hindbrain (Silva et al. 2021). Thus, NAM‐like microglia exist not only in the neurogenic area of the mid‐ and hindbrain but also in the retinal CMZ, raising the possibility that NAM supports more common mechanisms that maintain neural stem cells and their neurogenic processes in broader areas of zebrafish brain. In contrast to zebrafish, in mammals, including humans, neural stem cell‐mediated adult neurogenesis is maintained only in a limited brain area, such as hippocampus. It will be interesting to learn whether hippocampus‐resident microglia in mice or humans have transcriptome profiles similar to NAM. In research on zebrafish retinal neurogenesis, there has been a long debate on how much the neurogenic mechanism differs between post‐embryonic CMZ and embryonic retinas. Identification of 4w_Mg2 as retinal CMZ‐resident NAM suggests that 4w_Mg2 may be essential to support neurogenesis and neural circuit formation in the post‐embryonic CMZ in zebrafish. Another interesting possibility is that the retinal CMZ provides an essential route by which definitive hematopoiesis‐derived microglia colonize the retina and mature. In zebrafish, embryonic microglia derived from the rostral blood island (RBI) are progressively replaced with adult microglia derived from aorta‐gonad‐metanephros (AGM) (Xu et al. 2015; Yu et al. 2023). We previously found that RBI‐origin embryonic microglial precursors enter the optic cup along the hyaloid artery (HA) (Ranawat and Masai 2021). However, after closure of the choroid fissure, the HA route seems to be physically closed (Eckert et al. 2020). Instead, the superficial choroidal vessel system forms, extending veinous or arterial routes into the optic cup along the surface of retinal CMZ (Kaufman et al. 2015). Recently, it was also reported that in zebrafish, microglial precursors migrate into the optic cup through the basal surface of the retinal CMZ along the superficial ocular vasculature, especially the ventral radial vessel (VRV), in later development, after choroid fissure closure (Zhan et al. 2025). These data suggest that AGM‐origin adult microglial precursors migrate into the retina along the superficial ocular vasculature. We previously showed that microglia precursors infiltrate the retina through the neurogenic area in zebrafish at embryonic stages (Ranawat and Masai 2021). Since neurogenesis continues to be active in the retinal CMZ, microglial precursors may infiltrate the retina through the CMZ and differentiate into 4w_Mg2 cells. In the future, it will be important to verify this possibility.

In this study, we found that PLX3397 does not prevent Müller glia from entering cell proliferation, but keeps them or their‐derived neural progenitor cells from moving toward the ONL in pde6c mutants. Inflammation is required for Müller glia‐mediated neuronal regeneration in zebrafish (Iribarne and Hyde 2022; Silva et al. 2020; Zhang et al. 2020), and pro‐inflammatory cytokine, il1b, is necessary to activate Müller glia‐mediated neuronal regeneration in zebrafish (Lu and Hyde 2024). PLX3397 is an inhibitor of csf1ra and csf1rb, the absence of which affects survival of microglia (Oosterhof et al. 2018; Wu et al. 2018). So, it is interesting to ask why the PLX3397‐mediated effect differs from other microglial elimination methods such as dexamethasone. GOI expression analysis revealed that expression levels of csf1ra/csf1rb genes are variable, depending on microglial clusters, suggesting that susceptibility of PLX3397 may differ in each microglial cluster. Indeed, the number of microglia is not significantly reduced by PLX3397 treatment at 5 dpf. These surviving microglia are localized in various retinal areas including CMZ, RGC, INL and ONL, suggesting that they are likely 5d_Mg1. Expression of csf1ra and csf1rb genes is very low in 5d_Mg1, but highly expressed in both 5d_Mg2 and 5d_Mg3. Thus, it is likely that microglial clusters highly expressing csf1r genes are more sensitive to PLX3397 treatment. At 4 wpf, csf1r genes are highly expressed in 4w_Mg1, moderately expressed in 4w_Mg2 and 4w_Mg3, and absent in 4w_Mg4. Indeed, HCR‐FISH of PLX3397‐treated wild‐type retinas confirmed that the fraction of 4w_Mg1 is markedly reduced in PLX3397‐treated wild‐type siblings. Thus, PLX3397 treatment selectively eliminates 4w_Mg1. However, HCR‐FISH analysis showed that the fraction of 4w_Mg1 seems to be maintained in PLX3397‐treated pde6c mutants. GOI expression analysis revealed that expression of csf1rb, which regulates predominantly adult microglial development in zebrafish (Ferrero et al. 2021), is downregulated in pde6c mutants, which may explain survival of 4w_Mg1 in PLX3397‐treated pde6c mutants. Rather, our HCR‐FISH analysis showed that the fraction of “Other”, which contains non‐CMZ‐resident 4w_Mg2 and 4w_Mg3, is reduced in PLX3397‐treated pde6c mutants. The fraction of OS/RPE‐resident 4w_Mg2 is increased in pde6c mutants (Figure 7E), non‐CMZ‐resident 4w_Mg2 may promote Sox2⁺ Müller glial migration to the ONL. Alternatively, partial reduction of 4w_Mg3 may compromise Sox2⁺ Müller glial migration to the ONL. Further investigation will be necessary to clarify the role of microglia in Müller glial migration to the ONL.

Author Contributions

Conceptualization: DR, YT, IM; Methodology: DR, YT, IM; Software: DR, YT, IM; Validation: DR, YT, IM; Formal analysis: DR, YT, IM; Investigation: DR, YT, IM; Resources: DR, YT, IM; Data curation: DR, YT, IM; Writing – original draft: DR, IM; Writing – review and editing: DR, YT, IM; Visualization: DR, YT, IM; Supervision: IM; Project administration: DR, YT, IM; Funding acquisition: DR, IM.

Funding

This work was supported by a grant from the Okinawa Institute of Science and Technology Graduate University to IM, and Grant‐in‐aid for JSPS fellows (DC2), no. 22KJ3081 and no. 22J10161 (2022) to DR.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Microglia eliminate dying photoreceptors in pde6c mutants at 5 dpf.

Figure S2: Heatmap, clustering, and GO enrichment for 5dpf microglia.

Figure S3: HCR‐FISH for 5dpf and 4 wpf microglia.

Figure S4: Heatmap, clustering, and GO enrichment for4 wpf microglia.

Figure S5: 4w_Mg 2 and 4w_Mphage1 are similar to NAM and SAM, respectively.

Figure S6: mpeg1.1: EGFP⁺ cells located at the basal border of the CMZ express lygl mRNA, but not apoeb mRNA.

Figure S7: PLX3397 treatment of 5 wpf and 5 dpf zebrafish retinas.

Movie S1: Analysis of HCR‐FISH signal quantification in microglia.

Table S1: Key resource table.

Table S2: Probe sets, buffers, and hairpins for HCR‐FISH.

Table S3: Amplifier hairpin sets for HCR‐FISH negative controls.

Table S4: Criteria for microglial cluster assignment.

Table S5: Parameters for quality filtering of each object.

Dataset: 1 List of enriched genes for each microglial and macrophage cluster at 5 dpf.

Dataset: 2 List of enriched genes for each microglial and macrophage cluster at 4 wpf.

Dataset: 3 Summary of HCR‐FISH data for microglial cluster assignment in zebrafish retinas at 5 dpf and 4wpf as well as PLX3397‐ and DMSO‐treated retinas at 4–5 wpf.

Data Availability Statement

The data that support the findings of this study are openly available in NCBI's Gene Expression Omnibus (GEO) at https://www.ncbi.nlm.nih.gov/geo/, reference number GSE298347. This will be available after publication.