Fate mapping reveals that microglia and recruited monocyte-derived macrophages are definitively distinguishable by phenotype in the retina

The recent paradigm shift that microglia are yolk sac-derived, not hematopoietic-derived, is reshaping our knowledge about the isolated role of microglia in CNS diseases, including degenerative conditions of the retina. However, unraveling microglial-specific functions has been hindered by phenotypic overlap of microglia with monocyte-derived macrophages. The latter are differentiated from recruited monocytes in neuroinflammation, including retina. Here we demonstrate the use of fate mapping wherein microglia and monocyte-derived cells are endogenously labeled with different fluorescent reporters. Combining this method with 12-color flow cytometry, we show that these two populations are definitively distinguishable by phenotype in retina. We prove that retinal microglia have a unique CD45lo CD11clo F4/80lo I-A/I-E− signature, conserved in the steady state and during retinal injury. The latter was observed in the widely used light-induced retinal degeneration model and corroborated in other models, including whole-body irradiation/bone-marrow transplantation. The literature contains conflicting observations about whether microglia, including in the retina, increase expression of these markers in neuroinflammation. We show that monocyte-derived macrophages have elevated expression of these surface markers, not microglia. Our resolution of such phenotypic differences may serve as a robust way to help characterize isolated roles of these cells in retinal neuroinflammation and possibly elsewhere in CNS.

The breakthrough discovery of the embryonic origin of microglia in adult mice is redefining our understanding of these cells in immunopathogeneses of neuroinflammation and CNS conditions, including degenerative diseases of the retina. Ginhoux et al. formally demonstrated that microglia arise from yolk sac-primitive macrophages and persist into adulthood 1 . Adding to this finding were the important works of Ajami et al. 2 and Mildner et al. 3 , who both showed that circulating monocytes (Mo) do not contribute to the microglial pool under normal physiologic conditions, and that local self-renewal sustains microglia maintenance throughout adult life. Ajami et al. also showed that bone marrow-derived cells do not contribute to the pool of microglia even following Mo recruitment in inflammation 2-4 (a noted exception to this observation can occur in the setting of lethal irradiation/bone marrow transplantation 2,3 ). Considerable attention has been shifted toward redefining the isolated role of microglia, with the understanding that these cells are distinct from recruited Mo-derived cells in inflammatory diseases of the CNS. Early reports point to the possibility of non-redundant roles for these cells 5 , as implicated in models of Alzheimer's disease 6 , experimental autoimmune encephalomyelitis (EAE) 7 , and in models of retinal degenerative disease 8,9 . Nonetheless, discovering the isolated role of microglia in the brain and the retina has proven to be technically challenging. For one, microglia and certain Mo-derived cells have been deemed phenotypically indistinguishable. Indeed, recruited classical Mo can be resolved phenotypically because they express Ly6C and CCR2, but not F4/80 or Iba-1. However, differentiated Mo-derived macrophages (mo-MFs) overlap phenotypically with

Results
Phenotypic characterization of microglia in the normal retina of C57BL/6 mice. Our first aim was to comprehensively phenotype microglia in the normal retina using a multi-parameter flow cytometry panel of currently used myeloid cell markers. Markers included live/dead, CD45 (leukocyte), CD11c (dendritic cell; DC), I-A/I-E (antigen presenting cell), and F4/80 (macrophage; MF). We also incorporated CD64 (MF), also known as Fcγ RI, which was recently validated as a faithful marker for MFs 27 . In addition, markers such as CD11b (myeloid), Ly6C (Mo), and Ly6G (neutrophil; PMN) were also incorporated to discern Mo and PMN. In subsequent experiments, endogenous reporters (e.g. YFP, GFP, RFP) are also included. In all, a maximum of 12 parameters could be simultaneously analyzed using flow cytometry. In the current experiment, we analyzed normal retina from C57BL/6 adult mice to identify microglia, the predominant myeloid cell population in steady-state CNS parenchyma.
In the setting of whole-body lethal irradiation/bone marrow transplantation, microglia are phenotypically different than recruited mo-MFs. The next aim was to apply our flow cytometric strategy to determine whether recruited Mo-derived cells are phenotypically different from microglia. Our approach was to use whole body lethal-irradiation of host mice followed by donor GFP + bone marrow transplantation (Rad/GFP-BMT). Donor mice ubiquitously expressed GFP driven by the β -actin promoter. This approach offers several advantages. First, a direct effect of this procedure is low-grade chronic injury of the retina 21 which allows us to gain insight regarding myeloid phenotypes in this injury setting. Second, the procedure itself leads to CCL2-mediated recruitment of Mo into the retina 21 , allowing us to directly compare the phenotypes of GFP + (i.e. donor) Mo-derived cells with GFP − (i.e. host) microglia. C57BL/6 hosts were therefore subjected to whole-body Rad/GFP-BMT, and retinas were collected 3 months later. Data showed a significant population of extravascular myeloid (CD11b + ) cells in the retina that were GFP + (i.e. donor) at this time-point and a large population of remaining GFP − (i.e. host) microglia (Fig. 2a). With respect to intravascular GFP + classical Mo, their phenotype was IV-CD45 + CD11b + Ly6G − Ly6C hi CD64 − CD45 + CD11c lo F4/80 − I-A/I-E − (Fig. 2b). We were not able to detect an appreciable presence of extravasated GFP + classical Mo (data not shown), however, GFP + mo-MFs were readily detectable and their phenotype was IV-CD45 − CD11b + Ly6G − Ly6C − CD64 + CD45 + CD11c + F4/80 + I-A/I-E + (Fig. 2b). In contrast, GFP − microglia were IV-CD45 − CD11b + Ly6G − Ly6C − CD64 + CD45 lo CD11c lo F4/80 lo I-A/I-E − (Fig. 2b). These data revealed unexpected potential differences in expression levels of CD45, CD11c, F4/80 and I-A/I-E in microglia vs. mo-MFs, which was verified by comparing their mean fluorescence intensities (MFI) (Fig. 2c). MFI data of each individual sample was determined (Fig. S1). Thus, our data suggests that, in the setting of whole-body Rad/GFP-BMT, microglia have a unique CD45 lo CD11c lo F4/80 lo I-A/I-E − profile, and that their profile is not shared by recruited mo-MFs, which express significantly higher levels of these markers.
In light injury, microglia are phenotypically different from mo-MFs, as revealed in shielded Rad/ GFP-BMT hosts. Our next aim was to examine the phenotype of microglia and recruited mo-MFs in a more acute and robust injury setting. However, we first had to modify our whole-body Rad/GFP-BMT fate mapping technique because, as shown above, the whole-body approach causes retinal injury 21 and recruitment of mo-MFs. Therefore, we shielded the heads of host mice during lethal irradiation, as previously described 3 . We then validated that at 3 months post shielded Rad/GFP-BMT, host mice were protected from irradiation-induced recruitment of GFP + donor cells into the retina (Fig. 3a). As an alternative method of injury, we utilized a common light injury model wherein mice are exposed to toxic levels of bright white light that cause acute photoreceptor cell   (a) Head shielding protects host mice from radiation and radiation-induced Mo recruitment. C57BL/6 mice were irradiated with shielding then given donor GFP + bone marrow. Retinas were analyzed 3 months later. Cells were pre-gated on live CD45 + singlets. (b) Light injury leads to myeloid cell recruitment in shielded Rad/GFP-BMT hosts. Three months after shielded Rad/GFP-BMT, mice were then subjected to light challenge (or not) and retinas were harvested 7 days later for analysis. Cells were pregated on live CD45 + singlets. (c,d) Comprehensive phenotypic analysis reveals that microglia have a CD45 lo CD11c lo F4/80 lo I-A/I-E − profile during light injury, which is significantly different from recruited mo-MFs. GFP − microglia from retina of uninjured mice were compared to GFP + mo-MFs and IV Mo from retinas of light-injured mice. Significant differences of data shown (mean ± s.d.) were determined across samples by an unpaired t test (****p < 0.0001). Light injury data is representative of n = 4 individual samples; pre-light injury data is comprised of n = 2 individual samples that are representative of 2 independent experiments. Scientific RepoRts | 6:20636 | DOI: 10.1038/srep20636 death 30 . The advantages of this model include that injury occurs primarily at the level of the neuroretina (specifically the photoreceptors) 30 and that myeloid cells are significantly involved 9,31,32 . Thus, the following experiment applied light injury in shielded Rad/GFP-BMT hosts to enable us to discern newly recruited Mo-derived cells from microglia in the retina. The LED light injury model utilized here is described in greater detail in the methods sections.
Shown here, shielded Rad/GFP-BMT was performed and then 3 months later host mice were exposed to a light challenge (or left unchallenged) and retinas were collected 7 days later, a time-point consistent with substantial light-induced injury (data not shown). Agreeing with our previous experiments, we observed a significant population of extravascular donor-derived GFP + myeloid cells in light injury and a large population of host-derived GFP − microglia (Fig. 3b). For more in depth phenotypic examination, we compared mo-MFs and microglia from retinas of shielded Rad/GFP-BMT hosts, in uninjured vs. light-injured mice; specifically, we compared recruited GFP + myeloid cells from light-injured retinas vs. GFP − microglia from uninjured counterparts (Fig. 3a). The latter was necessary to avoid contamination of GFP − recruits in our analysis, as full chimerism in shielded Rad/GFP-BMT cannot be achieved (Fig. S2a). With respect to intravascular Mo, their phenotype was (Fig. 3c). Recruited mo-MFs were readily detectable and their phenotype was (Fig. 3c). Once again, we verified via MFI analysis that the expression levels of CD45, CD11c, F4/80 and I-A/I-E in microglia vs. mo-MFs were significantly different (Fig. 3d). MFI data of each sample was determined (Fig. S2b). Thus, as revealed by shielded Rad/GFP-BMT hosts, our data showed that in the light injury setting, microglia have a unique CD45 lo CD11c lo F4/80 lo I-A/I-E − profile. Furthermore this phenotype was not shared by recruited mo-MFs, which express significantly higher levels of these markers.
Fate mapping with CX 3 CR1 YFP−CreER/wt :R26 RFP hosts corroborates that microglia are phenotypically different from mo-MFs in light injury. Our next aim was to use an alternative approach for tracking myeloid cell origins to possibly corroborate that microglia have a unique CD45 lo CD11c lo F4/80 lo I-A/I-E − profile not shared with mo-MFs. We generated CX 3 CR1 YFP−CreER/wt :R26 RFP mice to utilize a previously described fate mapping strategy accomplished by pulsing mice with tamoxifen to label CX 3 CR1 expressing cells (i.e. YFP + cells) with RFP. Indeed, we found that 2 days after tamoxifen administration, YFP + circulating Mo and YFP + retinal microglia both expressed RFP (Fig. 4a). However, 60 days post tamoxifen administration, only retinal microglia maintained RFP expression (Fig. 4a); whereas RFP expression in circulating Mo was "washed out" by 60 days (Fig. 4a). Thus, at the 60-day time-point, retinal microglia and circulation Mo exhibited a YFP + RFP + and YFP + RFP − phenotype, respectively. Next, we exposed "washed out" mice to light challenge and analyzed retinas 5 days later. We observed a YFP + RFP + population consistent with microglia and a YFP + RFP − population consistent with recruited Mo-derived cells (Fig. 4b). In contrast, mice that were not exposed to light challenge only possessed YFP + RFP + microglia (Fig. 4b). Using flow cytometric enumeration, we observed that the total number of YFP + cells in the retina significantly doubled after injury (Fig. S3a). Interestingly, when we enumerated YFP + RFP + microglia vs. YFP + RFP − mo-MFs, there was no significant increase in microglia, suggesting that the observed increase in overall cellularity was entirely a result of newly recruited cells (Fig. S3a).
Next, we performed more in depth examinations of retinal cell phenotypes after light injury. With respect to intravascular Mo (YFP + RFP − ), their phenotype was IV-CD45 + CD11b + Ly6G − Ly6C hi CD64 − CD45 + CD11c lo F4/80 − I-A/I-E − (Fig. 4c). We also identified a small population of extravascular YFP + RFP − Ly6C + Mo-derived cells (Fig. S3b). However, these cells expressed F4/80 and CD64 (Fig. S3c), suggesting that their differentiation into mo-MFs was already underway. RFP − Ly6C − mo-MFs were readily detectible and their phenotype was (Fig. 4c). In contrast, RFP + microglia were IV-CD45 − CD11b + Ly6G − Ly6C − CD64 + CD45 lo CD11c lo F4/80 lo I-A/I-E − (Fig. 4c). Again, we were able to verify via MFI analysis that the expression levels of CD45, CD11c, F4/80 and I-A/I-E in microglia vs. mo-MFs were significantly different (Fig. 4d). MFI data of each individual sample was determined (Fig. S3d). Moreover, we were able to demonstrate that these populations can be discerned without fate mapping (Fig. 4e). Data shows gating of high and low expressing populations using CD45 alone, CD45 in combination with F4/80 or CD11c, and F4/80 in combination with CD11c. Based on these gates, we assessed the expression of RFP. Consistently, the gated population with lower CD45, F4/80, or CD11c contained the vast majority of RFP + microglia with minimal spillover of RFP-mo-MFs. Conversely, the gated population with higher CD45, F4/80, or CD11c contained the vast majority of RFP − mo-MFs with minimal spillover of microglia. Finally, we directly compared microglia from the steady state to those from light-injured retina, addressing whether any shifts in expression were detectable. Results indicate that changes in expression, if any, were marginal as compared to mo-MFs (Fig. S4a,b). Fluorescence minus one controls were included (Fig S4a,b).
Next, the ability to discern YFP + RFP + microglia and recruited YFP + RFP − Mo-derived cells in CX 3 CR1 YFP−CreER/wt : R26 RFP mice was also verified by confocal analysis (Fig. 5). Taken together, we revealed that in the light injury setting, microglia have a unique CD45 lo CD11c lo F4/80 lo I-A/I-E − profile which is not shared by recruited mo-MFs.
Lastly, we showed that our flow cytometric strategy is applicable to other mouse strains. Exposure to 60 k lux for 8 hr was required to damage the retina in mice on the C57Bl/6 background, e.g. CX 3 CR1 YFP−CreER/wt :R26 RFP mice (Fig. 5b). In contrast, mouse strains known to be susceptible to light injury, like BALB/c and CB6F1J 33,34 , required only 40 k lux for 4 hr (Fig. S6b). Similar to our results in CX 3 CR1 YFP−CreER/wt :R26 RFP , damaged CB6F1J mice showed invasion of myeloid cells to the outer retina (Fig. S6c). We performed more in depth examinations of retinal cell phenotypes after 5 days post light injury in CB6F1/J mice. Data show gating of high and low expressing populations using CD45 in combination with CD11c (Fig. S6d). The CD45 hi CD11c + population was not present in baseline retina, suggesting that it was a newly recruited population. We lacked fate mapping abilities in the CB6F1/J mice, but we analyzed the expression of F4/80 and I-A/I-E on the gated populations (Fig. S6e). Consistent with the phenotype of recruited Mo-derived cells, the CD45 hi CD11c hi population was also F4/80 + I-A/I-E + . And consistent with the phenotype of microglia, the CD45 lo CD11c lo population was F4/80 lo I-A/I-E − . Taken together our data suggest that microglia and Mo-derived cells have discernible phenotypes across mouse strains.

Analysis of extravascular I-A/I-E hi myeloid cells in normal retina.
Over the course of our experiments, we noticed a small but consistent subpopulation of extravascular I-A/I-E hi myeloid cells in the normal retina. We devoted the last series of experiments to further analyzing this subpopulation. A previous report by Lehmann et al. described a population of retinal cells in CD11c-DTR/GFP mice as having GFP and MHC II expression 35 . Thinking these cells might correspond to the I-A/I-E hi myeloid cells that we observed, we also surveyed retinas of CD11c-DTR/GFP mice. Our data clearly showed a small but distinct population of GFP + myeloid cells (Fig. S5a,b), which were specifically deleted following systemic diphtheria toxin (DT) administration , which is achieved by tamoxifen administration followed by a "wash out" period. Tissues were harvested 2 days after the last dose, or following a 60 day "wash out" period. Cells were pre-gated on live CD45 + CD11b + Ly6G − singlets. (b) Fate mapping discerns microglia from recruited Mo-derived cells in light injured retinas. Following tamoxifen pulsing and subsequent "wash out" period, mice were subjected to light challenge (or not) and retinas were harvested 5 days later. (c,d) RFP + microglia were compared to RFP − mo-MFs and IV Mo. Comprehensive phenotypic analysis reveals that microglia have a CD45 lo CD11c lo F4/80 lo I-A/I-E − profile during light injury, which is significantly different from recruited mo-MFs. Data from uninjured and light-injured retina are representative of n = 2 and n = 5 individual samples, respectively. Data is representative of two independent experiments. Significant differences were determined within samples by an unpaired t test (****p < 0.0001). (e) Microglia and mo-MFs are distinguishable by surface staining of CD45, F4/80, or CD11c. Gated on extravascular CD11b + retinal cells. Data shows different gating strategies using CD11b, CD45, and F4/80 to identify high and low expression cell populations. Subsequently, we analyzed RFP expression in each population to determine if the gating corresponds to genetic fate mapping. Cells were pre-gated on live CD45 + CD11b + singlets. Data is representative of at least 2 independent experiments; each experiment has n = 4 individual samples.
Next, we performed an in depth phenotypic analysis of the I-A/I-E hi subpopulation by directly comparing these cells with I-A/I-E − cells (microglia) (Fig. 6a). Extravascular myeloid cells from the retina of C57BL/6 mice were separately gated based on I-A/I-E hi or I-A/I-E − expression (Fig. 6a). Subsequent phenotypic analysis showed that the expression of several markers was increased in the I-A/I-E hi population (Fig. 6b), and thus appeared to be phenotypically distinct from I-A/I-E − microglia. We also noted the I-A/I-E hi population expressed CD64 and F4/80 (Fig. 6b), thus allowing us to designate these cells as putative MFs 27 .
To assess whether the I-A/I-E hi putative MFs might be long-lived and radio-resistant, like microglia, we subjected mice to either whole-body or shielded Rad/GFP-BMT. Three months later, the retinas of these mice were analyzed. We gated on IV-CD45 − CD11b + GFP − events (Fig. 6c), excluding GFP + myeloid recruits from our phenotypic analysis. Results showed that the presence of the I-A/I-E hi putative MFs following shielded Rad/ GFP-BMT hosts was abundantly clear (Fig. 6c), suggesting that these cells are long-lived and consistent with the MF phenotype. In contrast, the presence of these putative MFs following whole-body Rad/GFP-BMT was rather marginal (Fig. 6b), suggesting that these cells are not radio-resistant.
To further probe the possibility of a long-lived status for such I-A/I-E hi putative MFs, we employed our fate mapping technique with CX 3 CR1 YFP−CreER/wt :R26 RFP mice. Mice were tamoxifen pulsed and analyzed 3 months later after "wash out". We also included a cohort of mice that were light injured. After gating on RFP + (i.e. long-lived) cells, a distinct population of I-A/I-E hi cells was observed both in normal and in light-injured retina (Fig. 6d). Taken together, these data suggest that I-A/I-E hi putative MFs are long-lived, albeit not radio-resistant, whereas I-A/I-E − microglia are long-lived and radio-resistant.

Discussion
The current study addresses unresolved questions concerning the presence of bona fide microglia vs. recruited mo-MFs and their possible phenotypic differences in the injured retina. Our novel work involves the use of fate mapping with CX 3 CR1 YFP−CreER/wt :R26 RFP mice or bone marrow chimeras, in conjunction with multi-parameter flow cytometric analyses to address these questions. We established a technique that uses IV infusion of CD45 mAb to tag circulating leukocytes (i.e. Mo, PMN, etc.) within retinal vasculature, which facilitates subsequent exclusion of these cells by flow cytometry analysis. In combining such tools, our approach enabled us to directly demonstrate in retinal injury that microglia (i.e. embryonically-derived/tissue-resident) are distinct from recruited mo-MFs (adult bone marrow-derived/recruited). These two populations have only recently been accepted as ontogenically separate cell populations and are currently considered by many to be phenotypically indistinguishable. We show in the normal retina that microglia have a unique CD45 lo CD11c lo F4/80 lo I-A/I-E lo profile that is maintained in two different types of injury, chronic/low-grade retinal injury 21 (whole-body Rad/ BMT model) and acute/robust retinal injury (light challenge model 30 ). Thus, the CD45 lo CD11c lo F4/80 lo I-A/I-E lo signature distinguishes microglia from recruited mo-MFs with high fidelity and this phenotype may serve to be a helpful way in future studies to discern these two populations in retina and possibly elsewhere in the CNS by flow cytometry.
Our results regarding microglial phenotype in retinal injury from whole-body Rad/BMT or light-induced damage, challenge the dogma that these cells upregulate certain markers in the neuroinflammed setting. The complete phenotype of microglia in the retina demonstrated here, is as follows: IV-CD45 − CD11b + Ly6G − Ly6C − CX 3 CR1 + CD64 + CD45 lo CD11c lo F4/80 lo I-A/I-E − . This expression profile agrees with that of normal brain microglia, as demonstrated by Gautier et al. 27 . Strikingly, however, we found that their CD45 lo CD11c lo F4/80 lo I-A/I-E − expression is maintained following injury, even when we directly compared microglia from steady-state vs. injured retina. Thus, the previously held notion that activated microglia increase their expression of these markers in the photoreceptor degeneration setting 12,36-38 is in contrast with our light damage data. Also, we made the same finding in the whole-body Rad/BMT setting, which may suggest the presence of a generalized phenomenon. For example, microglia are thought to increase MHC II expression (referred to here as I-A/I-E) in models of EAE [39][40][41] , and in models of glaucoma and retinal CMV infection 42,43 . Likewise, microglia in EAE are thought to upregulate CD45 39,44 and CD11c 39 . The latter is also believed to increase on microglia in models of cuprizone injury 45 , and Alzheimer's 46 . However, our data suggest that myeloid cells with increased expression of such markers may instead be monocyte-derived. Our combined use of multi-color flow cytometry (including our IV-CD45 technique) with fate mapping strategies provided optimal resolution to discern microglia from extravasated mo-MFs. Our finding has functional implications as it is well documented that surface level expression of receptors impact responses (e.g. I-A/I-E levels often correlate with antigen presentation efficiency). Additionally, the fate mapping strategy demonstrated here will be an invaluable tool for future characterization of the isolated role(s) of microglia vs. mo-MFs in retinal diseases, such as certain forms of uveitis, photoreceptor degeneration, age-related macular degeneration, and diabetic retinopathy.
Another very interesting finding made here was the identification of a I-A/I-E hi myeloid population in the normal retina, which our data suggest are long-lived, but not a radio-resistant, MFs. Their long-lived characteristic was revealed by maintenance of RFP expression in CX 3 CR1 YFP−CreER/wt :R26 RFP mice, as well as in shielded Rad/ GFP-BMT hosts. A microglial designation for these cells would be in agreement with original flow cytometry studies by Sedgwick and colleagues, who identified a similar I-A/I-E hi population deemed microglia in brain and spinal chord of normal Brown Norway rats 41 . However, in our study, these I-A/I-E hi MFs had substantially elevated levels of F4/80 (and other markers), which we believe differentiates them from microglia. Furthermore, unlike microglia, these cells appeared to be radiosensitive. Supporting the notion that these cells are not microglia is an early report by Dick et al., who demonstrated an I-A/I-E hi population in the normal rat retina that co-expressed CD163 (i.e. ED2) 47 , a marker associated with perivascular MFs based on previous reports 48  Regarding recruited mo-MFs in retinal inflammation, our study demonstrates that these cells, not microglia, definitively express increased levels of CD45, CD11c, F4/80 and I-A/I-E following whole-body Rad/BMT or light challenge. Our IV-CD45 technique allowed us to resolve their likely precursors within the retinal vasculature, i.e. circulating classical monocytes (IV-CD45 + CD11b + Ly6G − Ly6C hi CX3CR1 + CD64 − CD45 + CD11c − F4/80 − I-A/I-E + ), from recruited mo-MFs (IV-CD45 − CD11b + Ly6G − Ly6C − CX 3 CR1 + CD64 + CD45 + CD11c + F4/80 + I-A/I-E + ). Our data also suggest that expression of CX 3 CR1 cannot be used to distinguish microglia and mo-MFs, as both expressed similar levels following light injury. We designated the new recruits as mo-MFs due to their strong F4/80 and CD64 expression. However, they also express CD11c, a marker traditionally associated with DCs but is now recognized to be expressed by certain MFs 27 . The CD11c expression on mo-MFs makes these recruits the likely equivalents to CD11c + recruits previously described in photoreceptor degeneration by others. For example, Chen et al. reported the presence of YFP+ cells in the retina of CD11c-YFP mice; a strain of mice that was showed to posses an RD8 mutation known to cause photoreceptor degeneration 50 . Another example includes GFP + cells in light-injured retinas of CD11c-DTR/GFP mice 35 . Interestingly, our data may reflect subtle variations in mo-MF phenotypes in response to injury. When contrasting CD11c expression by mo-MFs after light injury vs. whole-body Rad/GFP-BMT, we observed a trend for reduced CD11c expression after whole-body Rad/GFP-BMT, albeit still higher than microglial levels. The latter may be a reflection of chronic/low-grade inflammation evoked by whole-body Rad/GFP-BMT 21 , and indicates that there can be variability in recruited mo-MF phenotypes over time.
Continuing on the theme of recruited cells, our data may suggest the presence of an exceedingly rapid Mo to MF differentiation in the retina. Appreciation for this conceivable phenomenon was prompted from the rarity by which we were able to detect extravasated undifferentiated classical Mo (i.e. Ly6C hi F4/80 − ) following whole-body Rad/GFP-BMT or light challenge. Sparsely present extravascular Ly6C hi cells (also CCR2 + ) were instead found here to be F4/80 + and CD64 + , suggesting that these Mo-derived cells had already begun differentiation. It is tempting to speculate that such a rapid Mo to MF differentiation is a unique feature of inflammation in the retina or CNS as a whole. This concept would be supported in the work by London et al., who detected the presence of mo-MFs in the retina within 12 hr post glutamate intoxication 19 and Howe et al., who detected these cells in brain parenchyma within 18 hr post intracerebral TMEV inoculation 51 . In contrast to the CNS, in peripheral tissue inflammation, mo-MFs emerge approximately 72 hr after, as seen in models of experimental colitis 14 , myocardial infarction 52 or corneal wounding in mice 53 . In myocardial infarction and corneal wounding classical monocytes (i.e. Ly6C hi F4/80 − ) are detectable out to 7 days post injury, which was not the case in the retina following light injury in our study. Future work is required to validate this theory of an exceedingly rapid Mo to MF differentiation in the retina and to determine the functional relevance.
In summary, we directly demonstrate through fate mapping and bone marrow chimeras, in conjunction with 12 color flow cytometry, that microglia are distinct from recruited mo-MFs. Our data also identifies a population of long-lived I-A/I-E hi putative MFs in the steady state, which may implicate a perivascular MF designation. In addition, our data also led us to appreciate the possible existence of a profoundly rapid Mo to MF differentiation in the retina. Most importantly, we prove that retinal microglia have a unique CD45 lo CD11c lo F4/80 lo I-A/I-E − signature that is conserved in both steady state and during injury, which breaks from the traditional dogma that microglia increase expression of these markers in neuroinflammation. Instead, we show that it is the recruited mo-MFs that have elevated expression levels of CD45, CD11c, F4/80, and I-A/I-E. These microglia and mo-MFs have only recently been accepted as distinct and are considered by many to be phenotypically indistinguishable. However, the phenotypic differences shown here may serve as a way to henceforth characterize the isolated role/s of such cells in retinal diseases pathologies, such as uveitis, photoreceptor degeneration, age-related macular degeneration, and diabetic retinopathy. The β -actin GFP transgenic mice were generated as previously described 54 and were kindly provided by the G. H. Kelsoe Lab (Duke University). CD11c-DTR/GFP mice were kindly provided by the S.N. Abraham Lab (Duke University); mice were originally purchased from Jackson Laboratory [stock No. 004509]. All mice are housed at a barrier-free and specific-pathogen-free facility at Duke University School of Medicine (Durham, NC). All procedures were approved by the Institutional Animal Care and Use Committee at Duke University, and the procedures were carried out in accordance with the approved guidelines.

RPE65 PCR.
The PCR method to identify Methionine (Met) or Leucine (Leu) allele variations for the gene RPE65 was previously described 34 . We preformed the PCR on three different mouse strains: C57Bl/6J, BALB/c, and CB6F1/J (Fig. S6a) and found them to be Met/Met, Leu/Leu, and Met/Leu, respectively. The CX 3 CR1 CreER−YFP/CreER−YFP and R26 RFP strains were also confirmed to be RPE65 Met/Met (data not shown).
Light challenge. Light damage is induced with energy/cost-efficient white LED lamps, as previously described 55,56 . These generate 90% less heat than current fluorescent lamps and emit short-wavelengths overlapping with rhodopsin absorption. After 8 hours of dark-adaption, mouse eyes were dilated with atropine, 1% solution (Bausch & Lomb, Tamp, FL) and phenylephrine hydrochloride, 10% solution (Paragon BioTeck, INC., Portland, OR). Mice were then placed in a ventilated reflective container that contained available food and hydrogel. A cool white-light LED light source (Fancierstudio, San Francisco, CA) was placed above the container and the lux output was adjusted to 60 k using a luxometer. After 8 hrs of light challenge, the mice were returned to normal mouse housing facility lighting.
C57Bl/6 mice are known to be resistant to light injury due to a Leu to Met mutation in the REP65 gene 34 . Our use of 60 k lux for 8 hrs is required for injury of our transgenic mice since they are on a C57Bl/6 (Met/Met) background (Fig. 5b). Lower intensity exposures (40 k lux/4hrs) in C57Bl/6 mice do not result in retina injury or subretina transmigration of Iba-1 + cells (Fig. S6b,c). In contrast, mice that contain at least on copy of the Leu RPE65 allele are susceptible to light injury at lower intensity exposures, as shown in Fig. S6b in BALB/c (Leu/Leu) and CB6F1/J (Leu/Met) mice (Fig. S6b,c).

Labeling of intravascular CD45 cells.
Mice were anesthetized by i.p. injections of ketamine/xylazine (120 and 20 mg/kg, respectively). Once anesthetized mice were injected retroorbitally with 3.0 ug of biotin-labeled anti-CD45 (Biolegend, San Diego, CA). Mice were euthanized 5 minutes post injection to ensure labeling of blood cells.
Retina and blood harvest. Blood was collected from freshly euthanized mice via cardiac puncture. Cells were then treated with red blood cell lysis buffer (Sigma-Aldrich, St. Louis, MO) and thoroughly washed before immunostaining for flow cytometry. Eyes were gently enucleated. Using a dissection microscope, the cornea and lens were removed, exposing the retina. The retina was gently separated from the RPE using forceps and the optic nerve was severed at the back of the eye such that the retina could be isolated for digestion. Flow cytometry immunolabeling. Single cell suspensions of retina or lysed blood were transferred into PBS for staining with Aqua Live/Dead viability dye (Thermo Fisher Scientific, Waltham, MA) for 30 minutes then washed. Cells were incubated in a blocking solution containing 5% normal mouse serum, 5% normal rat serum, and 1% Fc Block (eBiosciences, San Diego, CA) and subsequently stained with a combination of fluorophore-conjugated primary antibodies against CD45, Ly6C, Ly6G, CD64, CD11c, CD11b (Biolegend), F4/80 (eBiosciences), I-A/I-E (BD Biosciences, San Jose, CA) and CCR2 (R&D Systems, Minneapolis, MN). Samples were stained at room temperature for 20 minutes; then a fluorophore-conjugated streptavidin (Biolegend) was added to each sample and staining continued for an additional 5 minutes. After completion of staining, cells were washed and fixed with 0.4% paraformaldehyde in PBS. Data was acquired with BD Fortessa flow cytometer using BD FACSDiva software (BD Biosciences). Raw flow cytometry data was analyzed using FlowJo software (FlowJo LLC, Ashland, OR) Histology and immunolabeling. After removal of the cornea and lens, the eyecup (choroid and neuroretina) were fixed in 2% paraformaldehyde in PBS for 3 hrs at room temperature. Tissues were successively washed with PBS, 15% sucrose, and then 30% sucrose before embedding in OCT and frozen. Frozen sections were cut at 40 um for immunostaining with anti-GFP/YFP (Invitrogen) and DAPI. Z-stack images were collected using confocal microscopy (Leica) and image analysis preformed on Imaris (Bitplane, Zurich, Switzerland).

Tamoxifen injections.
Tamoxifen (Sigma-Aldrich) was dissolved in corn oil to a stock concentration of 20 mg/ml. 75 mg/kg of tamoxifen was i.p. injected twice with one day in between injections. Mice were approximately 5-6 weeks of age when given tamoxifen.

Diphtheria toxin (DT) injections.
A dose of 0.5 ug of DT (Sigma-Aldrich) was administered i.p. two days prior to harvest.
Bone marrow chimeras. Chimeras were generated as previously described 57 . Recipient mice were lethally irradiated (1050 cGy). One cohort received whole-body exposure and another cohort was anesthetized then individually irradiated with lead shielding protecting their head. β -actin GFP transgenic donor mice were euthanatized and femurs and tibia collected. Distal and proximal ends of the bone were removed and the marrow flushed out with fresh RPMI medium. Single cell suspensions were thoroughly washed and resuspended in sterile HBSS. Immediately after irradiation, recipient mice received 5-10 million total GFP + bone marrow cells in 200 μ L via tail vein injection. Antibiotics were given in water to recipient mice.