Ocular macrophage origin and heterogeneity during steady state and experimental choroidal neovascularization

Background Neovascular age-related macular degeneration (nAMD) commonly causes vision loss from aberrant angiogenesis, termed choroidal neovascularization (CNV). Macrophages are heterogeneous cells that are necessary for experimental CNV, present in human CNV samples, and can display diverse functions, which are dependent upon both their origin and tissue microenvironment. Despite these associations, choroidal macrophage heterogeneity remains unexplored. Methods We performed multi-parameter flow cytometry on wildtype (WT) and Ccr2−/− mice after laser injury to identify macrophage subtypes, and determine which subsets originate from classical monocytes. To fate map tissue resident macrophages at steady state and after laser injury, we used the Cx3cr1CreER/+ ; Rosa26zsGFP/+ mouse model. We reanalyzed previously published single-cell RNA-seq of human choroid samples from healthy and nAMD patients to investigate human macrophage heterogeneity, disease association, and function. Results We identified 4 macrophage subsets in mice: microglia, MHCII+CD11c−, MHCII+CD11c+, and MHCII−. Microglia are tissue resident macrophages at steady state and unaffected by laser injury. At steady state, MHCII− macrophages are long lived, tissue resident macrophages, while MHCII+CD11c− and MHCII+CD11c+ macrophages are partially replenished from blood monocytes. After laser injury, MHCII+CD11c− macrophages are entirely derived from classical monocytes, MHCII− macrophages originate from classical monocytes (90%) and an expansion of tissue resident macrophages (10%), and MHCII+CD11c+ macrophages are derived from classical monocytes (70%), non-classical monocytes (10%), and an expansion of tissue resident macrophages (20%). Single-cell RNA-seq analysis of human choroid found 5 macrophage subsets: two MHCII+CD11C− and three MHCII+CD11C+ populations. One MHCII+CD11C+ subset was 78% derived from a patient with nAMD. Differential expression analysis identified up-regulation of pro-angiogenic gene expression in one MHCII+CD11C− and two MHCII+CD11C+ subsets, including the disease-associated cluster. The upregulated MHCII+CD11C− pro-angiogenic genes were unique compared to the increased MHCII+CD11C+ angiogenesis genes. Conclusions Macrophage origin impacts heterogeneity at steady state and after laser injury in mice. Both mice and human patients demonstrate similar macrophage subtypes. Two discrete pro-angiogenic macrophage populations exist in the human choroid. Targeting specific, pro-angiogenic macrophage subsets is a potential novel therapeutic for nAMD. Supplementary Information The online version contains supplementary material available at 10.1186/s12974-020-02010-0.

In this report, we investigated ocular macrophage heterogeneity, the origin of macrophage subsets, and their contribution to nAMD pathogenesis. Using multiparameter flow cytometry, we identified 4 macrophage subsets in mice: microglia, MHCII − , MHCII + CD11c − , and MHCII + CD11c + macrophages. Using the laserinduced CNV model, C-C chemokine receptor 2 (CCR2) knockout mice, and fate mapping studies, we demonstrated that each macrophage subset has distinct origins. After laser injury, MHCII − macrophages originate from classical monocytes and an expansion of tissue resident macrophages, MHCII + CD11c − macrophages are entirely derived from classical monocytes, and MHCII + CD11c + macrophages are derived from classical monocytes, nonclassical monocytes, and an expansion of tissue resident macrophages. At steady state, microglia and MHCII − macrophages are long lived, tissue resident macrophages, while MHCII + macrophages are partially replenished from blood monocytes. In order to translate these findings to human, we re-analyzed recently published singlecell RNA-seq data from the human choroid [14]. Similar to our findings in mice, we identified choroidal macrophage subsets categorized based on MHCII and CD11C expression, including two MHCII + CD11C − and three MHCII + CD11C + populations. Gene ontology (GO) enrichment analysis found three subsets enriched for angiogenesis-related genes. These studies demonstrate the presence of macrophage heterogeneity in mice and humans, and support the concept of pro-angiogenic, pathogenic macrophages, which could be future therapeutic targets.

Tamoxifen administration
Tamoxifen (T5648, Sigma-Aldrich, St. Louis, MO) was dissolved in corn oil (C8267, Sigma-Aldrich) at 20 mg/ ml with shaking at 37°C overnight. Tamoxifen solutions were stored at 4°C for less than 1 week. Tamoxifen was administered via intraperitoneal injection (75 mg/kg body weight) twice separated by 48 h via a 25G needle. Control animals received 100 μl of corn oil using the same technique. Tamoxifen or corn oil was administered at 6-8 weeks of age.

Laser-induced CNV
Male and female 10-12-week-old mice were treated as previously described [15]. Briefly, mice were anesthetized with ketamine/xylazine (Akorn, Lake Forest, IL). Pain control and hydration were achieved with a 1 mg/kg subcutaneous injection of Meloxicam (Henry Schein Animal Health, Melville, NY). Eyes were anesthetized and dilated, and a cover slip was coupled to the cornea with Gonak (Akorn) for slit lamp biomicroscopy and laser. Four (immunofluorescence) or eight (flow cytometry, to increase inflammatory cell numbers) focal burns (75 μm, 110 MW, 100 ms) were administered in each eye using a 532 nm argon ophthalmic laser (IRIDEX, Mountain View, CA) via a slit lamp (Zeiss, Oberkochen, Germany).

Flow cytometry of whole eyes
Experiments were performed as described [16]. Briefly, mice were sacrificed and eyes enucleated into HBSS.
Animals were not perfused as we previously demonstrated no difference in macrophage numbers at steady state or after laser injury with or without systemic perfusion [16]. Eyes were cleaned of optic nerve, extraocular muscles, orbital tissue, and conjunctiva. Whole mouse eyes including cornea, sclera, iris, ciliary body, vitreous, retina, and choroid, were minced into small pieces. Eye pieces were further mechanically and chemically digested before passing through a fine mesh filter to obtain a single cell suspension. Cell suspensions were stained for live cells and washed. Cell suspensions were blocked and stained with cell surface antibodies found in Table 1. Both eyes were pooled from one mouse to determine cells per mouse, using counts beads as previously described [16]. For dissected iris, choroid, and retina, only chemical digestion was performed without mincing of tissue or mechanical digestion. After passing the tissue through a fine mesh filter, the dissected and whole eye specimens were treated identically. Two unlasered mice in tamoxifen studies were pooled to increase macrophage numbers at steady state. Samples were run on a modified LSRII (BD Biosciences, San Jose, CA) and analyzed using FlowJo v10.

Flow cytometry of peripheral blood
Blood from sacrificed male animals was obtained with a 30G heparin needle via cardiac puncture. Samples were placed in EDTA tubes (Sarstedt, Numbrecht, Germany) to prevent clotting. In 5 ml polystyrene tubes, 1 μl of F c block was incubated with 90 μl of blood for 15 min at room temperature. To this mixture, 10 μl of an antibody cocktail (Table 1) was added. Samples were vortexed gently and incubated for 30 min at 4°C. Following incubation, 1.5 ml of lysis and fixation buffer (FACSLyse, BD Biosciences) was added. Samples were vortexed gently and placed for 10 min in the dark. Following lysis, 2 ml of MACS buffer (Miltenyi Biotec, Auburn, CA) stopped the reaction. Samples were centrifuged at 350×g at 4°C for 10 min. Pellets were resuspended in 500 μl MACS buffer and moved to 1.2 ml polypropylene tubes. This wash was repeated 3 times before resuspending in a final volume of 150 μl MACS buffer. Samples were run on a modified LSRII and analyzed using FlowJo v10.

Bioinformatics
Gene expression data (.tsv files) from human choroidal samples were downloaded from the GEO database (GSE135922). Data was imported into Seurat v3 [17,18]. The FindIntegrationAnchors followed by the Integrate-Data functions (dims 1:50) were used to integrate the data into one data set and perform batch corrections. The data were rescaled (ScaleData function), and principal component analysis (PCA) was performed (RunPCA, npcs = 50). The Elbow Plot technique was used to identify 19 significant principal components (PCs). Cells were clustered using FindNeighbors (dims = 1:19) followed by FindClusters (resolution = 0.4). The RunU-MAP function was used to visualize the cell clusters. Differential expression and cell identification were performed using FindAllMarkers (min.pct = 0.25, logfc.threshold = 0.25). The DotPlot function was used to visualize gene expression.
The leukocyte subset was created by making a subset of the leukocyte clusters, followed by scaling the data (Scale-Data), and PCA analysis (RunPCA). The Elbow Plot technique was again used to identify 12 significant PCs. The cells were clustered using FindNeighbors (dims = 1:12) followed by FindClusters (resolution = 0.4). Clusters were visualized using RunUMAP. Differential expression and cell identification were again performed using FindAll-Markers (min.pct = 0.25, logfc.threshold = log(2)). The DotPlot function was used to visualize gene expression. Enrichment in nAMD patients was performed by DimPlot (group.by=orig.ident) and table(eye.integrated@active.ident, eye.integrated$orig.ident). Gene expression was visualized using the VlnPlot function.
Gene ontology (GO) enrichment analysis was performed on upregulated and downregulated genes independently using a fold change cut-off = > 2.0 or < 0.5, and adjusted p value < 0.001. GOrilla was used for GO enrichment [19,20], using a background of genes expressed only in Mac-A, Mac-B, Mac-C, Mac-D, and Mac-E. REVIGO was used to remove dispensable GO terms [21]. All GO terms are visualized in Fig. 7 that met a dispensability cutoff of < 0.05, enrichment > 5-fold, number of genes (b) > 2, and false discovery rate (FDR) q value < 0.05.

Statistical analysis
Comparisons for CNV area were made by Mann-Whitney test due to non-parametric data distribution. Flow cytometry comparisons of macrophage numbers were made using the Brown-Forsythe and Welch ANOVA followed by Dunnett's T3 multiple comparison test due to unequal variances between unlasered and lasered mice.
The Ccr2 −/− mouse demonstrates deficient recruitment and mobilization of classical monocytes, and demonstrates reduced laser-induced CNV area [23], suggesting that classical monocyte-derived macrophages stimulate angiogenesis in the eye after laser injury. We used the Ccr2 −/− mouse model to investigate the influence of macrophage origin on each macrophage subtype. We first independently corroborated that female Ccr2 −/− mice demonstrate reduced CNV area at day 14 ( Fig. 2a, b). Alternatively, male Ccr2 −/− mice showed no difference in CNV area compared to wildtype (WT) male mice on day 14 (Fig. 2c). Based upon these sex differences and the increased prevalence of nAMD in female patients [24,25], we thoroughly investigated the number of macrophages after laser injury in male and female WT and Ccr2 −/− mice. We found that the CD45 dim CD64 + putative microglia, which are yolk sacderived, long-lived, and self-replenishing retinal macrophages, were unchanged by laser in both genotypes and sexes, as expected (Fig. 2d, e). In contrast, MHCII − macrophage numbers peaked with a 14.6-fold increase in female WT mice on day 3 (p < 0.01, Fig. 2f, g). In Ccr2 −/− female mice, MHCII − macrophage numbers increased from 257 ± 31 cells per mouse to 525 ± 67 cells per mouse (2.0-fold, p < 0.05 vs. day 0), which was significantly reduced compared to WT female mice on day 3 (p < 0.05, Fig. 2f, g). MHCII − macrophages were equally increased in female WT and Ccr2 −/− mice on day 5 (p < 0.05 for both genotypes vs. day 0), and back to baseline on day 7 (Fig. 2f, g).
MHCII + CD11c − macrophages increased 9.6-fold (p < 0.01 vs. day 0) on day 3 and 5.3-fold (p < 0.01 vs. day 0) on day 5 in female WT mice, and were completely unchanged in Ccr2 −/− female mice (Fig. 3a, b). In male mice, MHCII + CD11c − macrophages numbers elevated 5.8-fold (p < 0.001 vs. day 0) on day 3 and 2.7-fold (p < 0.01 vs. day 0) on day 7, and were completely unchanged in Ccr2 −/− male mice (Fig. 3c). These results suggest that the increase in MHCII + CD11c − macrophages with laser is entirely derived from classical monocytes. Alternatively, the laser injury-induced elevation of MHCII − and MHCII + CD11c + macrophages is primarily driven by classical monocytes, but there are additional contributions from non-classical monocytes and/or tissue resident macrophages in male and female mice.
In order to investigate tissue resident macrophages, we generated Cx3cr1 CreER/+ ; Rosa26 zsGFP/+ mice (Mac GFP , Fig. 4a). Upon tamoxifen administration, Cre expression in Cx3cr1 + cells will excise the stop codon prior to the zsGreen gene, causing irreversible bright zsGFP expression for fate mapping [26]. Mac GFP mice underwent intraperitoneal tamoxifen administration at 6 weeks of age. This strategy labeled Cx3cr1 + tissue resident macrophages and peripheral blood monocytes as GFP + . Peripheral blood monocytes are replenished from bone marrow after~2 weeks; thus,~4 weeks after tamoxifen administration, we expect peripheral blood monocytes to be GFP − . We performed multi-parameter flow cytometry on peripheral blood. One week after tamoxifen administration, peripheral blood monocytes (72.3%, Fig. S1E  S1I) demonstrated increased percentages of GFP + cells compared to corn oil vehicle controls. At 4-6 weeks post-tamoxifen treatment, monocytes and all other cell types had equivalent GFP + percentages compared to control mice ( Fig. S1E-J). These data confirmed that the Mac GFP mouse is an effective fate mapping model for tissue resident macrophages. Using our Mac GFP fate mapping model, we subjected male and female mice to tamoxifen administration at 6 weeks of age followed by laser treatment at 10-12 weeks of age (Fig. 4b). We performed multi-parameter flow cytometry on day 3 after laser injury, the peak of macrophage recruitment, in order to determine the contribution of tissue resident (GFP + ) vs. monocytederived (GFP − ) macrophages to each macrophage subset. The number of GFP + microglia increased with tamoxifen compared to corn oil control in female (Fig. 4d, p < 0.06) and male (Fig. 4h, p < 0.01) mice. Similarly, the number of GFP − microglia decreased with tamoxifen treatment in female (Fig. 4d, p < 0.06) and male (Fig. 4h, p < 0.01) mice. After laser injury, there was no change in GFP + microglia in male or female mice (Fig. 4d, h), which was expected because laser injury of the choroid should not significantly affect retinal microglia numbers. Alternatively, GFP + MHCII − macrophages increased from 49 ± 11 cells per mouse in the control to 96 ± 10 cells per mouse with tamoxifen ( Fig. 4e, p < 0.05) and 490 ± 109 cells per mouse with tamoxifen + laser (Fig. 4e, p < 0.05 vs. tamoxifen) in female mice. This effect size (4 00 cells) is equivalent to the increased number of macrophages observed in lasered Ccr2 −/− female mice on day 3 (Fig. 2g). GFP − MHCII − macrophages were dramatically upregulated from < 100 ± 6-11 cells per mouse in control and tamoxifen-treated mice to 4847 ± 1805 cells per mouse in tamoxifen + laser-treated mice (Fig. 4e, p < 0.09). Nearly identical findings were observed in male mice. GFP + MHCII − macrophages increased from 26 ± 1 in control male eyes to 48 ± 5 cells per mouse with tamoxifen ( Fig. 4i, p < 0.05) and 428 ± 93 cells per mouse in laser + tamoxifen-treated mice (Fig. 4i, p < 0.05). The increased GFP + MHCII − macrophage numbers were very similar to lasered Ccr2 −/− male mice, which increased from 169 ± 30 cells per mouse on day 0 to 397 ± 59 cells per mouse on day 3 (Fig. 2h, p < 0.05), suggesting that increased MHCII − macrophage numbers with laser are mainly due to classical monocyte recruitment with a minor contribution from expanded tissue resident macrophages.
We next queried the contribution of each patient to our 5 macrophage subsets to determine if any subset is disease-associated. We found that the macula of Donor 4, who had nAMD, represented 78.3% of cells in Mac-E (Fig. 6g, Table S3). No other cluster demonstrated > 50% of cells from one patient, suggesting that Mac-E could be nAMD-associated and potentially pro-angiogenic.
Although not as significant as the angiogenesis GO term in Mac-B, the Mac-D subset displayed 11.7-fold enrichment for regulation of endothelial cell migration (FDR q = 0.02, Fig. 7a). MicroRNA 29a (MIR29A) is expressed by macrophages and stimulates angiogenesis by downregulating anti-angiogenic genes like TIMP3 [33]. TIMP3 is the causative mutation for Sorsby's macular dystrophy, which demonstrates early onset CNV, and inhibits angiogenesis by blocking VEGF signaling [34]. MIR29A was increased 2.0-fold in Mac-D (Fig. 7i). Tumor necrosis factor (TNF) and platelet-derived growth factor B (PDGFB) are secreted ligands that stimulate angiogenesis in the laser-induced CNV model [35,36]. Both TNF and PDGFB were increased 2.2-and 2.1-fold in the Mac-D subset (Fig. 7j, k). Thus, Mac-D has the potential to be an additional pro-angiogenic population. Interestingly, no Mac-D upregulated genes in the regulation of endothelial cell migration GO term were shared with the Mac-B increased genes from the angiogenesis GO term.
Mac-E, the nAMD-associated population, was enriched 5.2-fold for cytokine-mediated signaling pathway (FDR q = 1.8 × 10 −8 , Fig. 7a). Due to our stringent dispensability cutoff, Mac-E enrichment for positive regulation of vasculature development (5.7-fold, FDR q = 6.6 × 10 −3 ) was subsumed by cytokine-mediated signaling pathway. MIR29A, which was the only gene unique to the positive regulation of vasculature development GO term, was upregulated 2.7-fold in Mac-E (Fig. 7i). TNF and PDGFB were included in both GO terms and increased by 2.6-and 3.3-fold respectively in Mac-E (Fig. 7j, k). In addition, the CXCR4 receptor, which is expressed on macrophages and is necessary for retinal and choroidal neovascularization [37], was expressed 2.5-fold more in Mac-E than other Mac subsets and included in both enriched GO terms (Fig. 7l). And finally, Mac-E demonstrated the highest VEGFA (2.7-fold, Fig. 7m) and lowest THBS1 (Thrombospondin-1, 3.2-fold downregulated, Fig. 7n) expression, the two central pro-and anti-angiogenic ocular genes in CNV [38], respectively. These analyses demonstrate similar macrophage heterogeneity between mice and humans, and support the concept of pro-angiogenic choroidal macrophage subsets, which could be future therapeutic targets for nAMD.

Discussion
In this report, we used the Ccr2 −/− mouse and the Mac GFP fate mapping model to investigate how macrophage origin impacts macrophage heterogeneity at steady state and during experimental CNV. At steady state, our data suggest a model where microglia and MHCII − macrophages are selfsustaining, tissue resident macrophages (Fig. 8a). In contrast, MHCII + macrophages are partially replenished by blood monocytes, with MHCII + CD11c + macrophages demonstrating a larger proportion derived from the monocyte pool (Fig.  8a). After injury, classical monocytes infiltrate the choroid and become MHCII − , MHCII + CD11c − , and MHCII + CD11c + macrophages (Fig. 8b). Additionally, tissue resident macrophages expand in response to injury and contribute to the MHCII − and MHCII + CD11c + macrophage populations. Finally, a small portion of MHCII + CD11c + macrophages are not accounted for in the Ccr2 −/− mouse or the Mac GFP model, suggesting potential involvement of non-classical monocytes.
Retinal microglia are the most well-studied macrophage in the eye, and are known to be long-lived, selfreplenishing tissue resident macrophages with no contribution from the monocyte pool [22]. Our data independently confirm this prior work at steady state (Fig.  5g). After laser injury, microglia are unchanged by laser or Ccr2-deficiency (Fig. 2d, e), suggesting that microglia numbers are relatively unaffected by laser injury to the (See figure on previous page.) Fig. 3 Ablated MHCII + CD11c − and decreased MHCII + CD11c + macrophage numbers in Ccr2 −/− mice in response to laser injury. a Representative flow cytometry pseudocolor plots from wildtype and Ccr2 −/− female mice from day 0, day 3, day 5, and day 7. MHCII + CD11c − macrophage numbers peaked in WT mice on day 3 and were completely unchanged in Ccr2 −/− female (b) and male (c) mice. MHCII + CD11c + macrophage numbers peaked in WT mice on day 3 and were blunted in Ccr2 −/− female (d) and male (e) mice. *p < 0.05; **p < 0.01; ***p < 0.001. Colored asterisk = significant difference from day 0 within genotype (blue = WT, red = Ccr2 −/− mice). Black asterisk = significant difference between genotypes on day 3, day 5, or day 7. N = 8-11 per group in both male and female mice. Macrophage numbers were compared using the Brown-Forsythe and Welch ANOVA followed by Dunnett's T3 multiple comparison choroid. An important limitation to our data is that because we distinguish microglia as CD45 dim CD64 + , activated microglia could not be included, which can express CD45 high CD64 + MHCII + markers. Those cells could be mischaracterized as MHCII + CD11c + macrophages, which could be an alternative explanation for the small expanded tissue resident GFP + MHCII + CD11c + population (Figs. 4g and Figs. 5k). How laser injury impacts microglia gene expression and function is less understood. Microglia are known to increase VEGF expression after laser injury [39], potentially implicating microglia as pro-angiogenic. Alternatively, depletion of CD11b + cells in the retina prior to laser injury, which depletes retinal microglia in addition to other cell types, has no effect upon CNV area [40], suggesting that microglia are not necessary for CNV compared to monocyte-derived macrophages. Because most prior studies do not sufficiently discriminate between microglia and infiltrating macrophages using multi-parameter flow cytometry or advanced fate mapping methods, the role of microglia during CNV remains unclear and more advanced methodologies will be needed to determine their function. Similar to retinal microglia and synovial macrophages [41], MHCII − macrophages are self-replicating with no replenishment from the peripheral monocyte pool at steady state (Fig. 5h). Since only 75% of MHCII − macrophages were GFP + , we suspect that this population is either Cx3cr1 low or heterogeneously Cx3cr1 + . Flow cytometry from ocular sub-compartments identifies that 60% of iris and ciliary body macrophages and < 20% of choroidal macrophages are MHCII − [42]. Our data similarly find that 70% of iris macrophages are MHCII − , but we find that 50% of choroidal macrophage are MHCII − (Fig. 5e). The difference between our data and the prior is both mouse background (Balb/c vs. C57BL6/J) and that we distinguish dendritic cells using CD64; since dendritic cells are MHCII + , their inclusion will decrease the fractional abundance of MHCII − macrophages in the prior study. Prior fate mapping studies show that iris and ciliary body macrophages are self-replenishing while choroidal macrophages are replenished from blood monocytes [13]. It is interesting that although 75% of MHCII − macrophages reside in the iris (Fig. 5d), all MHCII − macrophages are self-replenishing at steady state. This suggests that both cell surface markers and tissue microenvironment are equally important for determining cell fate.
After laser injury, MHCII − macrophages increase dramatically and are significantly reduced but not abolished in Ccr2 −/− mice (Fig. 2g, h). In Mac GFP mice, GFP + MHCII − macrophage numbers grew by~400 cells in male and female mice (Fig. 4e, i), which is nearly identical to the number of MHCII − macrophages found in Ccr2 −/− male and female mice after laser injury. These data suggest that after laser injury, MHCII − macrophages are 90% derived from classical monocytes and the remaining 10% expand from tissue resident macrophages (Fig. 8b). The fact that peripheral blood classical monocytes are~80% MHCII − and~20% MHCII + [43] supports our finding that the majority of MHCII − macrophages after laser injury are derived from classical monocytes.
We identified two MHCII + macrophage populations: MHCII + CD11c + and MHCII + CD11c − . At steady state, both MHCII + macrophage populations demonstrate replenishment from the peripheral monocyte pool (Fig. 5i, j). MHCII + CD11c − macrophages are 70% tissue resident with 30% derived from peripheral blood monocytes over a 2month period (Fig. 8a). MHCII + CD11c + macrophages show more frequent turnover, are 30% tissue resident, and 70% replenished from monocytes. CD11c is an alpha integrin that dimerizes with CD18 to form complement receptor 4 (CR4) [44]. The function of CR4 in macrophages includes adhesion to the vasculature and is important for tissue recruitment at steady state [45], potentially explaining why MHCII + CD11c + macrophages show more contribution from the monocyte pool. Additionally, MHCII + CD11c − macrophages displayed higher numbers in the iris compared to the choroid (Fig. 5d,  e). It is possible that the ocular sub-compartment is equally as important as CD11c expression in determining tissue resident vs monocyte-derived macrophage origin.
After laser injury, the increased MHCII + CD11c − macrophages numbers reflect an influx of classical monocytes differentiating into MHCII + CD11c − macrophages (Fig. 8b). This murine MHCII + CD11c − macrophage population is similar to the human Mac-A and Mac-B subsets (Fig. 6). Interestingly, Mac-B was 8.3-fold enriched for angiogenesis genes (Fig. 7b), including ligands (CYR61, ENG), receptors (TEK, LEPR), and transcription factors (ID1, EPAS1). Although Mac-B was not as over-represented as Mac-E, 37% of cells in Mac-B were derived from the (See figure on previous page.) Fig. 4 Fate mapping of macrophages subsets after laser injury. a Breeding strategy for generation of Mac GFP mice. b Experimental design for fate mapping of tissue resident macrophages. c Representative contour plots of GFP + vs. GFP − macrophage subsets. d-k Quantitative analysis of GFP + and GFP − macrophage numbers for each subset in female (d-g) and male (h-k) mice. Oil = corn oil vehicle control, Tam = Tamoxifen. *p < 0.05; **p < 0.01; ***p < 0.001. Color = GFP− (black) or GFP+ (green) cells. N = 3-9 in male mice and N = 4-9 in female mice per group. Macrophage numbers were compared using the Brown-Forsythe and Welch ANOVA followed by Dunnett's T3 multiple comparison macula of a patient with nAMD. Alternatively, as macrophage depletion causes choroidal vascular atrophy [10], Mac-B could support steady state choroidal vasculature homeostasis. Considering that MHCII + CD11c − macrophages are derived from classical monocytes, which drive angiogenesis in the choroid, and Mac-B, which is f Representative contour plots of GFP + vs. GFP − macrophage subsets. g Percent GFP + microglia were unchanged from week 1 to week 8. h Percent GFP + MHCII − macrophages were equivalent from week 1 to week 8. i Percent GFP + MHCII + CD11c − macrophages were decreased at week 4 and week 8. j Percent GFP + MHCII + CD11c + macrophages were decreased at week 4 and week 8. k Percent GFP + macrophage subsets at week 8 were significantly different. *p < 0.05; **p < 0.01; ***p < 0.001. N = 4-8 per group. Macrophage numbers were compared using the Brown-Forsythe and Welch ANOVA followed by Dunnett's T3 multiple comparison MHCII + CD11c − , is enriched for angiogenesis genes (Fig.  7b), specific anti-Mac-B therapy could be a novel therapeutic for nAMD.
After laser injury, MHCII + CD11c + macrophages are derived from classical monocytes, expanded tissue resident macrophages, and potentially from non-classical monocytes (Fig. 8b). Human scRNA-seq analysis identifies Mac-C, Mac-D, and Mac-E as MHCII + CD11C + macrophages (Fig. 6). These data demonstrate that MHCII + CD11c + macrophages are both heterogeneous in murine origin after laser injury, and heterogeneous in the human choroid. Interestingly, Mac-D was enriched for regulation of endothelial cell migration, and Mac-E demonstrated enrichment for cytokine-mediated signaling pathway, which included the positive regulation of vasculature development term (Fig. 7a). In support of a CD11c + disease-associated macrophage, CR4 is capable of binding Factor H [46] and is important for phagocytosis of iC3b opsonized particles [47]. Therefore, CD11c + macrophages are critical for complement-mediated processes (which are strongly linked to AMD). Furthermore, Mac-E was strongly derived (78.3% of cells) from a patient with nAMD, expressed the highest VEGFA, and lowest THBS1, which are two central angiogenic factors in nAMD. The subset of classical monocyte-derived MHCII + CD11c + macrophages are a potential key proangiogenic macrophage subtype.
Interestingly, the Mac-D and Mac-E upregulated genes for the regulation of endothelial cell migration and positive regulation of vasculature development GO terms were very similar to one another, but completely nonoverlapping with the angiogenesis genes from Mac-B. Therefore, our analysis has uncovered two discrete macrophage-driven angiogenesis functions in the choroid.
We identified a number of interesting sex-specific differences in the laser-induced CNV model. First, female Ccr2 −/− mice demonstrated reduced CNV area, while males did not (Fig. 2a-d). This is in partial agreement with prior reports that showed~75% inhibition of CNV in female Ccr2 −/− mice [23] compared to only 38% decreased CNV area in male Ccr2 −/− mice [48]. Similarly, female patients demonstrate increased prevalence of nAMD [24,25]. When observing absolute CNV area, this difference appears to be due to larger CNV area in WT female compared to male mice. Additionally, female WT mice on average displayed twice the number of macrophages compared to males (Figs. 2 and 3). However, male WT mice showed sustained increases in MHCII − , MHCII + CD11c − , and MHCII + CD11c + macrophage numbers on day 7, which was not observed in female mice. These data suggest that higher numbers of early macrophages in female mice caused larger CNV area, and/or prolonged macrophage presence decreased CNV area in male mice.
Our studies have a few important limitations. First, we used whole eye, including cornea, iris, ciliary body, vitreous, retina, and choroid for our studies. We made this decision for optimal rigor and reproducibility, and to prevent uneven ocular dissections from creating variance in our results. Because the site of injury is the choroid with some perturbations of the overlying retina, our increased macrophage numbers after laser injury most likely reflect choroidal and subretinal macrophages. However, we cannot exclude contributions from other ocular compartments. Secondly, the Cx3cr1 CreER transgene does demonstrate some expression without tamoxifen treatment, which is consistent with prior reports [49]. We made key comparisons between corn oil vehicle control and tamoxifen treatment, and demonstrated equivalent GFP + blood monocytes 4-6 weeks after tamoxifen treatment compared to vehicle control (Fig. S1), in order to minimize noise created by the leaky Cx3cr1 CreER transgene. Nevertheless, rare non-specific GFP + cells cannot be excluded. Lastly, we compared the number of Ccr2 −/− macrophages on day 3 vs. day 0 to the number of increased GFP + macrophages on day 3 after laser treatment. We found equivalent numbers for MHCII − macrophages from male and female mice, but less GFP + compared to Ccr2 −/− macrophages for the MHCII + CD11c + subset in male and female mice. From this data, we concluded that all MHCII − macrophages were either derived from classical monocytes or tissue resident macrophages, and that MHCII + CD11c + macrophages were derived from classical monocytes, tissue resident macrophages, and a third source that is either non-classical monocytes or activated microglia. However, these are two independent experiments, which may not be directly comparable, and these interpretations need to be confirmed with mice specifically deficient in non-classical monocytes.

Conclusions
In summary, we demonstrated the presence of MHCII − , MHCII + CD11c − , and MHCII + CD11c + macrophages in mouse eyes. Using the laser-induced CNV model, Ccr2 −/− mice, and Mac GFP fate mapping studies, we showed that MHCII + CD11cmacrophages were entirely derived from classical monocytes, MHCII − macrophages originated from classical monocytes and an expansion of tissue resident macrophages, and MHCII + CD11c + macrophages were derived from classical monocytes, potential non-classical monocytes, and an expansion of tissue resident macrophages. In human choroidal macrophages, we identified MHCII + CD11C − and MHCII + CD11C + populations. Two discrete macrophage-driven pro-angiogenic clusters were found in the MHCII + CD11C − and MHCII + CD11C + groups. These studies demonstrate that macrophage heterogeneity exists in the mouse and human choroid, that pro-angiogenic macrophages exist, and are a potential new therapeutic target for nAMD. Fig. 8 Model of ocular macrophage lineage during steady state and laser injury. a At steady state, microglia (red) and MHCIImacrophages (green) are tissue resident, self-sustaining, and long-lived macrophages. MHCII+ macrophages (CD11c − [gray] and CD11c + [blue]) are partially self-sustaining and partially recycled from the peripheral blood monocyte pool (numbers indicate percentage replenished from blood monocytes). b After laser injury, MHCII − macrophages (green) are derived from classical monocytes (90%) and from expanded tissue resident macrophages (10%). MHCII + CD11c − macrophages (gray) are 100% derived from classical monocytes. MHCII + CD11c + macrophages are derived from classical monocytes (70%), non-classical monocytes (20%), and expanded tissue resident macrophages (10%)