Structural and functional distinctions of co-resident microglia and monocyte-derived macrophages after retinal degeneration

Mice were cared for and handled in accordance with the National Institutes of Health guidelines and under protocols approved by the UC Davis Institutional Animal Care and Use Committee. All mice were born and maintained in constant darkness before exposure to constant light (approximately 200 lx) and were 2–5 months old when used for experiments, except for lineage mice which were treated with tamoxifen at 6–12 weeks and used for experiments at 4–9 months of age (after fluorescent cell markers were established). All mouse strains were tested to confirm the absence of the rd8 mutation. Although there are some reports of potential sex differences in microglial function [36], there were no observable sex differences in immunohistochemistry, flow cytometry, or in vivo imaging experiments, so both male and female mice were used for these experiments. Only female mice were used for single-cell mRNA sequencing to avoid potential confounding sex-specific differences in gene expression.

Arr1^−/− mice [2, 38] are a convenient, light-inducible model of photoreceptor degeneration. C57BL/6J mice (strain 000664), originally obtained from The Jackson Laboratory and then bred in constant dark conditions, were used as controls for some experiments. Cx3cr1-GFP mice (strain 005582, aka B6.129P2(Cg)-Cx3cr1^tm1Litt/J)[8] were obtained from The Jackson Laboratory and crossed with the Arr1^−/− strain to obtain Arr1^−/−Cx3cr1^+/GFP mice. Littermate Arr1^+/+Cx3cr1^+/GFP mice were used as controls. For lineage tracing experiments, Ai9 (strain 007909, aka B6.Cg-Gt(ROSA)26Sor^{tm9(CAG−tdTomato)Hze}/J)[16] and Cx3cr1-YFP-CreER (strain 021160, aka B6.129P2(Cg)-Cx3cr1^{tm2.1(cre/ERT2)Litt}/WganJ)[13] mice were obtained from The Jackson Laboratory and crossed with the Arr1^−/− strain to obtain Arr1^{+/+ or −/−} Ai9^KI/KI Cx3cr1^{+/YFP−CreER} mice for lineage tracing experiments.

A custom-built dual scanning laser ophthalmoscopy (SLO) and optical coherence tomography (OCT) system was used to simultaneously collect structural and fluorescent information from the retina non-invasively [40]. For imaging, mice were anesthetized with 2–2.5% isoflurane and kept on a 37 °C heating pad. A micropositioner (Biptigen, Morrisville, NC) was used for fine rotational and translational movements of the pad to position the mouse. Pupils were dilated with tropicamide and phenylephrine (Akorn), and then hypromellose gel (GenTeal Tears Severe, Alcon) was applied to the corneal surface. This gel both prevented the cornea from drying during imaging and maintained the refractive index between the cornea and contact lens (Unicon Corporation). SLO imaging was performed with a 488-nm laser, approximately 100–130 μW at the pupil, to excite GFP. Wide-field images were collected over 51° of visual angle, and increased digital sampling was used to “zoom” in on areas of interest. SLO images were processed and analyzed in FIJI [26], including registering images with the MultiStackReg plugin (a modified version of the TurboReg plugin [34] written by Brad Busse), averaging frames, and pseudo-coloring images for presentation.

OCT imaging was performed with a superluminescent diode centered at 860 nm with a 132-nm bandwidth (Broadlighter T-860-HP, Superlum), approximately 600 μW at the pupil. The FD-OCT data were processed using standard methods to transform the back-scattered light into intensity by depth information and then generate a series of OCT B-scans. As in our previous study [19], a custom Python script was used to flatten the B-scans and then register those flattened B-scans using a strip-registration algorithm to generate complete flattened OCT volumes. As previously described [19], focal light damage was performed by maintaining the OCT laser at a single location at full power (7–8 mW) for 2 min.

Intraperitoneal (IP) injections were used to deliver 5-ethynyl-2’-deoxyuridine (EdU) and tamoxifen for proliferation and lineage tracing experiments, respectively. For the proliferation experiments, EdU was dissolved in sterile 1 × phosphate buffered saline (PBS) at a concentration of 10 mg/mL and delivered at a dose of 100 mg/kg (or 10 ml/kg of the EdU solution). EdU was administered to dark-reared Arr1^−/− and C57BL/6J mice, approximately 2–4 months of age, at 48 and 72 h after the onset of light exposure. These are the time points that previous data [10] indicated microglia and macrophages are mitotically active. For the lineage tracing experiments, tamoxifen was dissolved in corn oil at a concentration of 20 mg/ml. Arr1^{+/+ or −/−} Ai9^KI/KI Cx3cr1^{+/CreER−YFP} mice received two IP injections of tamoxifen at a dosage of 75 mg /kg (or 3.67 ml/kg of the corn oil solution), separated by 1 day as previously described [20]. These IP injections were performed when the mice were approximately 6–12 weeks of age, and dark lighting conditions were maintained during injections. Further experiments and light exposure did not proceed until at least 60 days after the tamoxifen injections to allow all the monocytes to turn over, leaving only long-lived resident macrophages dual-labeled.

Immunohistochemistry was performed as previously described [24]. Immediately after enucleation eyes were submerged in 4% paraformaldehyde at room temperature. After approximately 5 min of fixation, anterior segments and lenses were removed, the resulting eyecups were fixed for an additional 20–25 min at room temperature, and then fixed eyecups were transferred to vials of PBS at 4 °C. To remove the retinal flat mount from the eyecup, several curvature-releasing cuts were made with small dissection scissors, and then small forceps were used to gently peel the retina from the rest of the eyecup. For antibody staining, retinas were incubated in 1% Triton X-100 in PBS overnight at 4 °C, blocked with normal serum for 2 h at 37 °C, incubated in primary antibodies in PBS with 0.5% bovine serum albumin and 0.5% Triton X-100 (PBT) overnight at 4 °C, rinsed 3 times in PBT for 10 min at room temperature, incubated in secondary antibodies for 1.5–2 h at 37 °C, rinsed in PBT again at room temperature, and then mounted on glass slides using ProLong Diamond Antifade Mountant (Invitrogen). For proliferation (EdU) experiments retinal flat mounts were permeabilized in 1% Triton X-100 for 45 min at room temperature, followed by using the Click-iT EdU Alexafluor 488 Imaging Assay (C10337, Invitrogen) to label EdU according to manufacturer instructions. Next retinal flatmounts were blocked with normal serum and then incubated with pre-conjugated antibodies, both for approximately 2 h at 37 °C. Finally, these flatmounts were rinsed 2 times in PBT for 10 min, once in PBS for 10 min, and mounted as above. See Table 1 for a full list of antibodies and labeling reagents used for immunohistochemistry. All retinal flatmounts were imaged using a Nikon A1 confocal microscope.

Maximum intensity projection images were created using NIS-Elements software (Nikon). ImageJ [26] was used to create the pseudocolored-overlay images shown in Figs. 4A and 6B; method depicted in Additional file 3: Fig. S3. Single-channel images were first thresholded using Color Threshold, then converted to a binary mask and overlayed using the Image Calculator set to Average. For YFP images with large amounts of background, speckle noise was removed using Remove Outliers (1–3 pixels) prior to processing. Grayscale overlay images from ImageJ were taken to Photoshop (Adobe), where cells were manually pseudocolored peach if dual-positive or blue if single-positive. To quantify the response to secondary focal damage in Fig. 6C, we counted the number of pixels of each color within the injury area. For the Sholl analysis in Fig. 5, maximum intensity projections were exported from NIS-Elements and binary images of individual cells were generated in ImageJ. An ImageJ plugin was used to perform the Sholl analysis [5], with a soma radius of 20 pixels and using 10 pixel steps. The number of primary branches, required for the calculation of the ramification index, was identified manually. For Fig. 5, cells with detectable levels of MHCII immunostaining were categorized as MHCII^high while undetectable levels were categorized as MHCII^low since we could not be sure based solely on IHC that cells were MHCII negative.

Retinal samples were prepared similarly to a previously described protocol [10]. Approximately 5 min before euthanasia mice received a tail vein injection of anti-mouse CD45 antibody conjugated to either PE or APC, to differentiate cells within the vasculature from those in the retinal parenchyma [21]. Mice were then euthanized, enucleated, and then retinas isolated. Isolated retinas were placed in 1 mL of digestion buffer, consisting of Hibernate medium (HBSS, 10-547F, Lonza) supplemented with 5% fetal bovine serum (FBS), 10 mM HEPES, 0.7 mg/mL calcium chloride, 1.5 mg/mL collagenase A (Roche), and 0.1 mg/mL DNase I (Roche). Tissue was incubated at 37 °C for 15–20 min, with mild trituration for all samples ultimately used for single-cell sequencing, or gently disrupted with a MACS Octo Dissociator (Miltenyi Biotec) for all samples used for flow cytometry. After digestion, the cell suspension was filtered through a 70-µm cell strainer (Genesee Scientific) and resuspended in PBS. Cells were then stained for viability (Zombie NIR Fixable Viability Kit, 423106, BioLegend), blocked for 5 min with Fc block 1 µl/sample (eBioscience) supplemented with 5% each normal rat and mouse sera, and then incubated with antibodies diluted in Brilliant Stain Buffer (BD Biosciences) for 15–30 min at room temperature. Table 1 includes a complete list of the antibodies and cell stains used for these experiments. After staining, cell suspensions were washed in 0.5% bovine serum albumin (BSA) with 1:50 EDTA, and ultimately resuspended in either the same buffer for cell sorting experiments or Cell Staining Buffer (BioLegend) for flow cytometry experiments. Additionally, flow cytometry samples were then fixed by adding 0.5% paraformaldehyde to the final cell suspensions. Control beads were prepared using the AbC Total Antibody Compensation Bead Kit (Invitrogen) and ArC Amine Reactive Compensation Bead Kit (Invitrogen) following manufacturer guidelines. Blood samples taken and processed in parallel as before [10] were used to set the CD45-IV gate for each experiment.

All flow cytometry and cell sorting was performed at the UCD Flow Cytometry Shared Resource Laboratory. For the time course flow cytometry experiment, samples were run on an LSR-II Cytometer (Becton Dickinson). For quantifying cells after lineage tracing, samples were run on a Cytoflex Cytometer (Beckman Coulter). Cell sorting prior to single-cell mRNA sequencing was performed on an Astrios EQ Cell Sorter (Beckman Coulter). FlowJO software (Tree Star) was used for analyses.

In all figures, data are represented as mean ± standard error and significance levels are indicated as follows: *p < 0.05, **p < 0.01, ***p < 0.001. All statistical analyses were performed using R version 4.0.0 or higher (R Core Team, n.d.). Two group comparisons (as in Figs. 1D and 5D) were performed using t-tests. Multiple group comparisons of immunohistochemistry and flow cytometry data (as in Figs. 2F–H and 5B–E) were performed using a one-way ANOVA followed by a Tukey’s honest significant differences test. Multiple group comparisons over time (as in Figs. 4C and 6C) were performed using a two-way ANOVA followed by a Tukey’s honest significant differences test. When any flow cytometry data were highly positively skewed, the data were first logarithmically transformed to improve the normality of the distribution and then statistics were run on the transformed data. Transformed data were never directly compared to non-transformed data, and only non-transformed data are shown.

Single-cell mRNA sequencing was performed as previously described [24]. Briefly, retinal samples were enriched for immune cells by FAC-sorting single, alive, CD45⁺CD11b⁺ cells, as described above. Cells were sorted into a solution of Dulbecco’s modified Eagle medium (DMEM; Thermo Fisher) supplemented with 10% FBS, and then immediately used to create the expression libraries. These libraries were generated using the 10X Genomics system (Chromium) at the UC Davis DNA Technologies and Expression Core, following manufacturer recommendations. The resulting cDNA libraries were sequenced on a NovaSeq (Illumina), again following manufacturer recommendations.

After sequencing, initial processing including barcode processing, transcript alignment, and generation of gene–barcode matrices was performed using Cell Ranger (Chromium) at the UC Davis Bioinformatics Core. All further processing and analysis was performed in R (R Core Team, n.d.), primarily using the Seurat R package [1]. Following the recommended Seurat workflow, cells were filtered for quality control, and then normalized for UMI and mitochondrial counts. After principal components analysis (PCA), we performed tSNE dimensional reduction and used marker gene expression to identify general cell types. Despite the FAC sorting, there was a small amount of contamination of non-immune cells, including photoreceptors. These cells were excluded, and then PCA, tSNE reduction, and clustering were performed on the remaining myeloid cells. Transcriptionally distinct clusters of cells were identified using Seurat’s graph-based clustering algorithm, and then cell types were assigned based on the relative expression of standard marker genes. Differentially expressed genes were identified with the FindMarkers function in Seurat using the Wilcoxon rank sum test, and only genes with an adjusted p-value below 0.05 were considered significantly differentially expressed. The cross-correlational analysis function in Seurat was used to combine this dataset with the previously published data from [24]. Gene enrichment analysis was performed using the Gene Ontology Panther tool [18].

During photoreceptor degeneration, microglia adopt an ameboid morphology and migrate to the outer nuclear layer [30, 39]. In the Arr1^−/− mouse model of light-induced photoreceptor degeneration, an influx of monocytes from the periphery and the proliferation of both the resident microglia and invading monocyte-derived macrophages contribute to a dramatic increase in the number of macrophages in the degenerating retina [9, 10]. To understand how this immune response resolves, we began by tracking Cx3CR1-GFP⁺ cells in the retina over time in vivo using Cx3cr1^+/GFP mice, which express GFP in both microglia and monocyte-derived cells. Dark-reared Arr1^−/−Cx3cr1^+/GFP and control Arr1^+/+Cx3cr1^+/GFP mice were exposed to light, then periodically imaged using scanning laser ophthalmoscopy (SLO) to track the distribution and morphology of GFP⁺ cells, and optical coherence tomography (OCT) to follow the concurrent degeneration (Fig. 1A). As previously reported [10, 24], we observed a dramatic increase in the number of GFP⁺ cells in Arr1^−/− retinas during photoreceptor degeneration (Day 2), and those cells were mostly ameboid in morphology. By 7 days photoreceptor loss was largely complete as revealed by the disappearance of the outer nuclear layer in OCT. At later times (Day 20), there appeared to be a fewer GFP⁺ cells compared to Day 7 and the morphology of many cells were again noticeably more complex and ramified (Fig. 1A).

To visualize the cell morphology at higher resolution, we amplified the GFP signal using immunohistochemistry (IHC) and performed confocal imaging (Fig. 1B, green). In retinal flatmounts, the GFP⁺ cells were ramified with complex secondary and tertiary processes in both WT and KO retinas. This is in stark contrast with the ameboid shape that these cells adopt during photoreceptor cell loss (for example, see [24]). However, the cells in the KO retinas had noticeable differences from the WT microglia mosaic, including more variable degrees of morphological complexity. Additionally, some cells expressed MHCII proteins (Fig. 1B, orange), which are expressed in monocyte-derived macrophages and sometimes activated microglia. Virtually no such MHCII⁺ cells were detectable in WT retinas processed in parallel.

Previously, we found that subpopulations of both microglia and monocyte-derived macrophages in the retina proliferate during the first few days of photoreceptor degeneration [24]. To determine if dividing cells survive and contribute to the retinal macrophage population after degeneration is complete, we pulsed C57BL/6J and Arr1^−/− mice with EdU at days 2 and 3 of light exposure, when cell proliferation has been previously detected in the Arr1^−/− model. Subsequently, on Day 20 eyes were collected and flat mounted retinas were stained for CD11b and EdU and visualized on a confocal microscope. Many EdU⁺ CD11b⁺ cells were observed in all layers of degenerated Arr1^−/− retinas, and only rarely observed in control (C57BL/6J) retinas (Fig. 1C). Cell counting from the histological images revealed approximately 45% of the CD11b⁺ cells were also EdU⁺ in the 20-day retinas (Fig. 1D). These results indicate that just under half of the immune cells in this newly homeostatic state were born during the intense period of photoreceptor degeneration roughly 2 weeks prior.

Although SLO imaging revealed fewer GFP⁺ cells at 20 days than at 7 days, there were still qualitatively more GFP⁺ cells in the retina at this late time compared to healthy controls (Fig. 1A, compare Day 0 to Day 20). To further quantify myeloid cells before, during, and after photoreceptor loss we used flow cytometry. We began by using general myeloid cell gating, identifying alive, single, CD11b⁺, CD45⁺ cells, and then further divided these cells by their degree of CD45 expression (see Fig. 2A for gating). Cells with relatively low CD45 expression initially increased in number, peaking at approximately 4 days of light exposure, and then gradually returned to baseline numbers (Fig. 2B). Low CD45 expression is often a feature of tissue-resident macrophages, so we further gated these cells using a standard microglia paradigm including Cx3cr1⁺, Ly6c^low, and MHCII^low expression. These microglia followed the same temporal pattern as the larger CD45^low gate, initially increasing and then returning to baseline cell numbers (Fig. 2C). Myeloid cells with higher CD45 expression also increased in number during the first few days of degeneration and then later decreased (Fig. 2D). Because high CD45 expression is typically a feature of peripherally derived cells, we further differentiated these CD45^high cells by their relative Ly6C expression, which is typically high in newly born monocytes. Cells that were Ly6C^high increased in number early in degeneration, peaked during the first day of light exposure, and then decreased to baseline numbers within approximately one week (Fig. 2E, red). In contrast, the Ly6C^low population peaked a few days later than the Ly6C^high, and never fully returned to baseline cell numbers (Fig. 2E, blue). These results suggest that the persistent increase in Cx3CR1⁺ cells that we observed after degeneration had waned (Fig. 1A) was due in part to continued presence of monocyte-derived macrophages. Thus, unlike the largely homogenous retinal macrophage population in healthy retinas, the retinal immune cell population remains heterogeneous long after the loss of photoreceptors.

To investigate the degree of heterogeneity of the new resident population, we performed single-cell sequencing on FACS-enriched retinal immune cells before, immediately following, and well past photoreceptor loss in Arr1^−/− retinas (0, 7, and 20 days, respectively). At all three time points, retinal CD45⁺ cells showed broad transcriptional similarities (Fig. 3A), suggesting that by 1 week many cells had returned to a “resting” state similar to microglia in a healthy retina. To further probe for heterogeneity, we identified transcriptionally distinct clusters of cells using Seurat’s graph-based clustering algorithm (Fig. 3B), and then evaluated expression of known marker genes (see Fig. 3C for examples) to determine the putative identities of these clusters. This analysis identified several distinct subpopulations, including resting microglia, mildly activated microglia, and a small number of inflammatory macrophages (Fig. 3B). Interestingly, the 7 and 20-day samples contained a unique cluster, Cluster #3, of hybrid-like cells expressing some genes traditionally associated with only microglia and specific disease-associated activated microglia [11, 31], such as Trem2, Apoe, Fabp5, and Spp1, as well as genes traditionally associated only with cells of a monocytic lineage, such as Adgre1, Cd74, and H2-Aa (Fig. 3C).

When we probed the expression of common pro- and anti-inflammatory related genes, the only cluster with significant expression of either was the cluster containing a very small number of inflammatory macrophages (Cluster #4). All other clusters lacked strong expression of both pro- and anti-inflammatory markers. This suggests that, although Clusters #2 and #3 are distinct from traditional “resting” microglia, these macrophages are not strongly activated. This is especially apparent when examining canonical resting microglia markers, like Hexb, P2ry12, and Siglech, and activated microglial markers, such as Lyz2, Sepp1, and Apoe (Fig. 3C). Specifically, in Clusters #2 and #3 resting microglia marker expression was slightly lower on average and more variable, while the activated markers were somewhat higher than in Cluster #1.

The similarities between the mildly activated microglia in Cluster #2 and the putative monocyte-derived macrophages in Cluster #3 prompted us to look for differentially expressed gene programs between these groups. We identified the top 35 most highly differentially expressed genes in the monocytic cells compared to the microglia and performed gene enrichment analysis (Additional file 1: Fig. S1A). The most enriched gene programs included macrophage activation, regulation of cell migration, antigen processing and presentation, and response to interferon-gamma. These enriched gene sets align with a monocyte-derived macrophage phenotype. We also performed gene enrichment analysis on the top 400 most highly expressed genes in the monocytic cluster, which yielded similar results (Additional file 1: Fig. S1B).

Finally, we combined this dataset with a previously published dataset from this model at an earlier timepoint during active degeneration (t = 2 days; [24]). A tSNE plot was used to graphically compare the similarities between clusters identified in these two experiments (Fig. 3D). The control (0 days) resting microglia clusters from these two datasets largely overlapped. Especially notable in this plot is that Cluster #3 is located between the monocytic cells from the earlier timepoints, and the activated microglia from both datasets. This is in agreement with the seemingly hybrid phenotype of both intermediate microglial and monocytic gene expression in this cluster.

Given the expression of both microglial and monocyte-derived macrophage genes in the hybrid-like cells of Cluster #3, we hypothesized that many monocyte-derived cells were adopting a microglia-like phenotype and remaining in the retina long-term. To test this hypothesis, we used an inducible fate-mapping paradigm to track the lineages of retinal immune cells after degeneration. In this lineage tracing paradigm, all Cx3cr1-expressing myeloid cells, including microglia and monocyte-derived cells, express YFP and a tamoxifen-inducible Cre (a Cre-ERT2 fusion protein, referred to here as CreER). Upon tamoxifen administration, Cre-mediated recombination of the Ai9 reporter results in tdTomato expression in all Cx3cr1-expressing cells. Then, 60 + days after tamoxifen, by which time all the monocytes have turned over, long-lived resident macrophages, including bona fide microglia, are YFP⁺ and tdTomato⁺, whereas the newly born cells from a monocytic lineage are only YFP⁺ [20].

To examine the lineage of the cells remaining in the retina after photoreceptor degeneration had ended, we administered the tamoxifen prior to the onset of any degeneration, in both WT and Arr1^−/− dark-reared animals. After nine weeks, well after the required 60-day period for monocyte turnover, we exposed both strains to light (~ 200 lux) to initiate degeneration. Following degeneration (20 days), we examined both control and Arr1^−/− retinas for dual- and single- labeled macrophages (Fig. 4A). In control retinas (WT, Arr1^+/+Ai9^KI/KI Cx3cr1^{+/YFP−CreER} post-tamoxifen), all of the retinal microglia were dual-labeled, as expected. In contrast, in the degenerated retinas (KO, Arr1^−/−Ai9^KI/KI Cx3cr1^{+/YFP−CreER} post-tamoxifen), single-labeled cells were abundant. These cells were distributed across all retinal layers and appeared intermixed with dual-labeled cells within the macrophage mosaic. These results demonstrate that following degeneration, infiltrated monocyte-derived cells take up long-term residence in the retina. Furthermore, these single-labeled cells had ramified morphologies, supporting the idea that these monocyte-derived macrophages adopted a microglia-like phenotype (Fig. 4A).

To quantify the relative numbers of both resident (CD45⁺YFP⁺tdTomato⁺) and monocyte-derived (CD45⁺YFP⁺tdTomato⁻) myeloid cells, we performed flow cytometry before and after photoreceptor loss (for gating, see Fig. 4B). We found that in the healthy retina, virtually all myeloid cells were resident cells (YFP⁺tdTomato⁺) (Fig. 4C, 0-day time point). However, by 7 days the dominant myeloid cell type was peripherally derived (YFP⁺tdTomato⁻), with the number of resident cells unchanged from the initial baseline (Fig. 4C, 7-day timepoint). By 20 days of light exposure, the populations of resident and peripherally derived macrophages were equivalent in size (Fig. 4C, 20-day timepoint). These results held even when the counts were normalized to the total number of recorded cells or the number of CD45⁺ cells (Additional file 2: Fig. S2).

To further compare the two lineages of macrophages, we quantified the morphologies of these cells in histological flatmounts using Sholl analysis. Resident microglia (YFP⁺tdTomato⁺) in control (WT, Arr1^+/+, 20 days light) and degenerated (KO, Arr1^−/−, 20 days light) retinas, as well as monocyte-derived macrophages (YFP⁺tdTomato⁻) in degenerated retinas were examined. Consistently, the resident microglia were larger and more complex than the peripherally derived macrophages. On average, microglia covered a larger spatial extent than monocyte-derived macrophages, as indicated by intersecting more radii (Fig. 5A), and had more complex morphologies, as indicated by more total intersections and larger ramification index, than monocyte-derived macrophages (Fig. 5B and C, respectively). Interestingly, the microglia after degeneration covered a smaller area (Fig. 5A) and were slightly less complex (Fig. 5B, C) than microglia in healthy retinas. Like the transcriptomic results above, this morphological analysis supports the conclusion that resident microglia after degeneration maintained a slightly activated phenotype.

Because the single-cell sequencing data (Fig. 3) suggested that monocyte-derived macrophages typically express higher levels of MHCII-related genes than microglia, we next examined MHCII expression in retinal flatmounts from the lineage tracing paradigm. As before (Fig. 1B), MHCII expression was not highly expressed in microglia of WT retinas. In contrast, in degenerated KO retinas (20 days of light) 60% of all monocyte-derived (YFP⁺tdTomato⁻) macrophages and 11% of all resident (YFP⁺tdTomato⁺) cells expressed high levels of MHCII (Fig. 5D, E). Sholl analysis revealed that resident cells that were MHCII^low had normal morphologies similar to those of WT retinas, and that all monocyte-derived cells and MHCII^high resident cells were smaller (Fig. 5F) and less complex (Fig. 5G, H) than normal. In fact, there were no statistical differences between the MHCII^high resident cells and monocytic groups. The only statistical difference between the MHCII^low resident groups is that after degeneration the cells are slightly smaller on average (Fig. 5F, WT vs KO; ANOVA followed by Tukey’s honest significant difference test, p = 0.0038). These results suggest that MHCII staining is a reasonable screening marker for monocyte-derived cells in the retina, although it should not be used as a conclusive identifier because it is likely to also reveal roughly 10% of the resident cells, which could include perivascular macrophages as well as microglia.

Given that the population of retinal CD45⁺ cells are heterogeneous after degeneration and that many cells have a mildly activated transcriptional profile, we next tested the ability of these immune cells to respond to further retinal injury. After photoreceptor degeneration was complete (20 days following light onset), we performed a focal laser injury in Arr1^−/−Cx3CR1^+/gfp mice and examined the retinal structure and GFP⁺ immune cells using OCT and SLO, respectively. Similar to the normal microglial response to focal laser damage [19], GFP⁺ cells in the degenerated retina migrated to and clustered at the focal site of damage within a few days. This cluster of cells then gradually disappeared and the GFP⁺ cells again spatially tiled the retina over a period of over a week (Fig. 6A). These results demonstrate that the populations of myeloid cells in the retina after degeneration can respond to further insult.

To determine whether these responding immune cells consist of resident microglia, monocyte-derived macrophages, or both, we performed focal light damage in the fate-mapping (Arr1^−/−Ai9^KI/KI Cx3cr1^{+/YFP−CreER}) mice after photoreceptor loss was complete and used immunohistochemistry to identify monocytic (YFP⁺tdTomato⁻) and microglial (YFP⁺tdTomato⁺) cells around the focal injury. Both microglia (YFP⁺tdTomato⁺) and monocyte-derived macrophages (YFP⁺tdTomato⁻) were visible in the tight cluster of cells that formed directly at the injury locus in the inner retinal and subretinal layers (Fig. 6B; see also Additional file 3: Fig. S3, Additional file 4: Fig. S4, Additional file 5: Fig. S5, Additional file 6: Fig. S6, Additional file 7: Fig. S7, Additional file 8: Fig. S8), but only the resident microglia showed a significant response to the secondary focal damage. Figure 6C shows that following focal damage (Days 1–4), there was a sharp increase in microglia in both the inner and outer retinal layers at the injury locus, indicating that the neighboring microglia (YFP⁺tdTomato⁺) more readily migrated than the monocyte-derived cells (YFP⁺tdTomato⁻). In the subretinal space, it was particularly noticeable that most of the amoeboid macrophages were resident (YFP⁺tdTomato⁺) in origin. By 14 days, the cluster had resolved and myeloid cells again retiled the retina in both layers. Considering that monocyte-derived macrophages comprise approximately half of the retinal macrophages at the timepoint focal damage was performed (Fig. 4C) and that the majority of macrophages at the damage location were microglial in origin (Fig. 6C), we conclude that monocyte-derived cells residing in the retina are substantially less responsive than resident microglia to further insults.

In previous work, the large increase in myeloid cells in the retina during photoreceptor degeneration in the Arr1^−/− mouse appeared to be the result of both infiltration of monocytes from the periphery and proliferation of myeloid cells within the retina [9]. Here we show that the number of myeloid cells eventually falls as degeneration wanes, but never quite returns to normal pre-degeneration levels. Because loss of photoreceptors during degeneration results in a dramatic decrease in the volume and total number of cells in the retina, even this relatively small increase in total immune cells results in a considerable increase in the density of immune surveillance within the retina. Our laser damage experiments show that this higher density of surveillance does not seem to affect the time course of the microglial response to subsequent injury compared to previous work in otherwise healthy retinas [19]. Whether or not the change in cell number or composition accounts for the apparently poor responsiveness of the monocyte-derived macrophages remains an open question. Interestingly, our EdU experiments show that roughly half of the myeloid cells in the post-degeneration mosaic were born when degeneration was most active (Fig. 1D), but we do not know whether these cells all locally proliferated or to what extent they entered the retina as EdU + newly born monocytes. Our lineage tracing experiments show that roughly half of the immune cells are monocyte-derived (Fig. 4C), but whether or not an equal fraction of microglia and monocyte-derived cells proliferated to re-populate the retina remains a question for future studies.

Our flow cytometry data here, in combination with previous work, suggest that these monocyte-derived macrophages arise from monocytes that invaded the retina during the first few days of degeneration, rather than resulting from continued invasion of monocytes throughout the entire course of degeneration. Newly born monocytes express high levels of Ly6C, which then decreases over the cell’s lifetime, especially if the monocyte differentiates into a monocyte-derived macrophage [29]. In previous work, flow cytometry results showed that the number of Ly6C^high cells in the Arr1^−/− retina peaked between 24 and 48 h during the height of photoreceptor degeneration, which corresponded well to the time course of the increase in Müller-cell derived CCL2 expression (monocyte chemoattractant protein) within the retina [10]. In this study, we examined the number of Ly6C^high cells in the retina at longer times, after degeneration had slowed and found that the number of Ly6C^high returned to non-detectable baseline levels by 10 days (Fig. 2E). Immunohistochemistry and Sholl analysis revealed that these monocyte-derived cells had processes, though were somewhat less ramified than the resident microglia (Fig. 5A–C). Thus although the integrity of the blood retinal barrier has not been explicitly examined in this model, together these results suggest that the monocyte invasion is limited to the peak period of degeneration, and that these monocytes rapidly differentiate into microglia-like macrophages that persist in the retina for at least several more weeks.

The monocyte-derived macrophages that take up residence in the retina after photoreceptor degeneration adopt a ramified morphology that is spatially intermingled with the endogenous microglia (Fig. 4A), similar to the monocyte-derived cells that infiltrate the retina following NaIO₃ administration to damage the RPE [15]. Unlike the macrophages in other ocular compartments with high turnover rates [37], these monocyte-derived macrophages in the retina are long-lived, and our morphological analysis shows that they are smaller and less complex that resident microglia on average (Fig. 5A–C). The extent of morphological difference between resident and monocyte-derived macrophages may vary by disease. For example, in the NaIO₃ model of RPE damage, resident and monocyte-derived macrophages were morphologically indistinguishable [15].

Although monocyte-derived macrophages adopt a largely microglial phenotype significant differences in gene expression remain. One such difference readily detectable at the protein level is MHCII. Immunohistochemically, the MHCII⁺ cells observed in the degenerated retina (e.g., Figure 1B) mostly correspond to the monocytic cells (ratio of 6:1 monocyte to microglia-derived cells; Fig. 5D). MHCII⁺ resident macrophages have been observed occasionally in other models of retinal/RPE injury [15, 37]. Notably, here we find that the relatively few resident cells that express high levels of MHCII proteins are morphologically indistinguishable from monocyte-derived macrophages (Fig. 5F–H). Thus, while MHCII staining may be a useful tool for qualitative assessment of peripheral infiltration that is much easier and cheaper than fate-mapping experiments, we emphasize that, like morphology, MHCII expression alone is not sufficient to determine the lineage of a given cell.

Monocyte-derived, or other non-microglial macrophages, have been observed to similarly adopt microglia-like phenotypes and remain in the retina in the absence of continuing or active neuronal loss in previous work. However, those instances have been situations in which there is physical damage [22], disruption of blood–tissue barriers [15, 37], or a pharmacological intervention that ablates the resident microglia [6]. Monocytic cells have been detected in other photoreceptor degenerative disorders, including widespread light damage [20] and age-related macular degeneration [27], although it has not yet been investigated if monocytes permanently engraft into the retina in these instances. To our knowledge, this is the first demonstration of monocyte-derived macrophages taking up long-lived residence in the retina following cell autonomous degeneration. The “new resident’’ monocytic cells adopt a microglia-like morphology and gene enrichment program aligned with a monocyte-derived macrophage phenotype; however, they fail to functionally respond to a subsequent insult as well as the resident microglia. The molecular basis of the apparent functional laxity of the monocyte-derived cells remains to be determined.

In the Arr1^−/− experiments presented here, monocyte-derived macrophages adopted a microglia-like phenotype over the course of days, consistent with evidence from other studies that the retinal environment promotes a microglia-like phenotype for myeloid cells. For example, after pharmaceutical ablation of retinal microglia, non-microglial immune cells from the periphery invade and adopt morphologies and gene expression reminiscent of microglia [6, 17, 22]. Non-microglial repopulation following microglia ablation has not been observed in the brain thus far [7]. However, following whole body irradiation and bone marrow or hematopoietic stem cell transplantation, non-microglial macrophages can invade the brain and develop similar microglia-like morphologies and some shared gene expression [28]. This study identified a non-microglia-specific gene expression program expressed following LPS challenge [28], and some of these genes were found to be similarly expressed in the small cluster of inflammatory macrophages identified here (Cluster #4), but are notably absent in the more microglia-like cells (Cluster #3, Fig. 3C). All together the CNS seems to promote a microglia-like phenotype for mononuclear phagocytes, although it remains unclear to what extent this phenomenon may differ between the brain and retina.

A specific disease-associated microglia (DAM) phenotype first observed at the single-cell level in a mouse model of Alzheimer’s disease [11] has been observed in a wide range of neurodegenerative disorders [4] and similar changes in microglial gene expression have been observed in several models of retinal degeneration [21, 24, 37]. Many DAM-associated genes were indeed expressed by activated microglia in this study, and interestingly these genes were also expressed in the microglia-like macrophages cluster (Fig. 3C). This phenotype of activation was not observed in the small cluster of pro-inflammatory monocyte-derived macrophages, indicating that this is not a pan-activation phenotype in this model.

Unexpectedly, many of the most highly expressed genes in the monocyte-derived macrophage cluster in our single-cell dataset have recently been identified as key genes expressed by various macrophage subsets in peripheral CNS tissues, such as the meninges and choroid plexus [35]. However, to what extent these conserved gene expression programs are indicative of shared functions, similar environment, or similar cell lineages remains unknown.

In the field of retinal degeneration, activation and resolution of the immune response has been investigated in a model of RPE injury that causes photoreceptor loss [15] and following a corneal alkali chemical burn that leads to retinal ganglion cell death [22]. In both models, expression of genes associated with immune activation are detected well after cell loss ceases, suggesting that at least some of the retinal immune cells do not return to rest [15, 22]. Because neither study utilized single-cell transcriptomics, the degree of heterogeneity in those post-degeneration immune populations remains unclear. However, the resolution of the immune response to neuronal loss has been investigated at a single-cell level in a model of facial nerve axotomy [32, 33]. Although this model does not result in widespread neuronal loss like the retinal Arr1^−/− model, the loss of neurons does drive the activation and proliferation of nearby microglia, and the immune response resolves approximately 30 days after injury. In this case, single-cell sequencing revealed that although the majority of microglia return to a resting phenotype, a small number of microglia continue to express many activation-related genes like the retinal degeneration models described above, and similar to the mildly activated microglia phenotype we observe in the single-cell dataset described here (Fig. 3).

Curiously, this persistent, mildly activated phenotype observed in all the above models has similarities to changes observed in microglia during aging. In the aging retina, microglia accumulate in the subretinal space, increase in number, and increase their expression of activation- and inflammation-related genes [3, 14]. Typically, aged microglia are not as responsive to injury as young microglia [14], yet here we find that microglia are still capable of robustly responding to focal injury even after all photoreceptors are lost (Fig. 6B, C). The similarities of microglia in neurodegenerative disorders and those in aging are noticeable and could have implications for further immune challenges or therapeutic interventions with age, especially after potential repeated challenges throughout life.

In summary, we find that photoreceptor degeneration causes long-lasting changes to the retinal immune environment, lasting well past when neuronal loss has completed. Although we find that these retinal macrophages can still respond to further retinal insult, the retinal immune environment does not entirely return to rest, and the microglial and monocytic cells respond differently to subsequent injuries and insults. Determining the precise molecular signals and pathways involved in these distinct subpopulations and their functional long-term differences, if any, will be the focus of future studies. Of particular interest will be determining if this altered immune state contributes to late-stage retinal remodeling, and if these immune cells affect therapeutic interventions such as gene therapies and cell transplants.