Brain Parenchymal and Extraparenchymal Macrophages in Development, Homeostasis, and Disease

Microglia, the parenchymal macrophages of the CNS, originate from cKit⁺CD41⁺CD45^lo erythromyeloid precursor cells (EMPs) generated during embryonic primitive hematopoiesis (1, 2) (Fig. 1). EMPs bud from clusters of Tie2⁺CD34⁺Flk1⁺ hemogenic endothelial cells in the yolk sac (YS) as early as Embryonic day 7.5 (E7.5) and provide the first cohort of macrophages and erythrocytes to the developing embryo (3–5). EMP-derived macrophage precursors (CD45⁺CD115⁺F4/80⁺CX3CR1⁺) appear in the blood around E8.5, spread via the embryonic circulation, and seed all the tissues including the CNS, wherein they complete the differentiation process (3, 6, 7). EMP commitment toward macrophages is supported by a combination of transcription factors, especially Runx1, PU.1, and IRF8 (4, 6, 8–10). Notably, microglia do not require the transcription factor Myb (11), which is necessary for the bone marrow hematopoiesis (12). YS-derived EMPs also migrate into the fetal liver (FL), thus giving rise to the transient-definitive hematopoiesis. During the third week of gestation, EMPs in the liver are progressively outcompeted by hematopoietic stem cells, which eventually establish definitive hematopoiesis in the bone marrow (13). Microglia first appear in the neuroepithelium at E9.5 (6, 8) (Fig. 1). However, invasion of the embryonic CNS ends with the formation of the blood–brain barrier (BBB) around E14.5 (14). Thereafter, microglia spread throughout the CNS parenchyma and undertake a stepwise differentiation process, reaching a final maturation state around the second postnatal week (15, 16).

The concept that mouse microglia are solely derived from YS EMPs dominates the field and is currently supported by several lines of evidence (3, 6, 17). However, a recent publication suggested that ∼20–25% of brain microglia stem from Hoxb8-positive FL-derived monocytes (FL-Mo), arising from FL hematopoiesis around E12 (18). Thus, FL-Mo would enter the embryonic CNS and acquire phenotype and functions barely distinguishable from YS-derived microglia. Albeit highly provocative, this model might explain previous findings that contradict a sole YS origin for microglia. First, a transient increase in Ly-6C⁺ monocytes was noted in the embryonic brain between E12 and E14 (19). Second, early depletion of YS-derived macrophages prevents the seeding of microglia in the brain rudiment at E10.5 and enhances the recruitment of monocytes at E14.5. Next, a microglial population reappears during the late gestational period, when EMPs are no longer present (19, 20). Third, the S100a4^Cre line, supposedly specific for monocytes, engenders fluorescent labeling in roughly 20% of embryonic microglia, and this population is maintained in the adult brain (19, 21). Further studies are needed to determine whether multiple hematopoietic waves contribute to microglia in mice. Current fate-mapping tools may not accurately resolve partially overlapping populations such as YS-derived macrophages and FL-Mo. In the near future, techniques involving photoinducible labeling, such as NICHE-seq (22), might be exploited to track the fate of macrophage-progenitors in vivo.

Microglial survival is dependent on the CSF1R ligands CSF1 and IL-34 (23–27). In the brain, IL-34 is produced by neurons, whereas CSF1 is produced by both neurons (26, 28, 29) and microglia themselves (15). Interestingly, these cytokines exhibit nonoverlapping expression patterns. CSF1 is highly expressed in the cerebellum, corpus callosum and spinal cord, whereas IL-34 is more abundant in the neocortex, olfactory bulb, striatum, and hippocampus (25–27). Genetic ablation of either CSF1 (24, 26) or IL-34 (25, 27) results in a partial reduction of microglia, estimated to be ∼30 and 50%, respectively. Furthermore, it has recently been shown that deletion of CSF1 in the neuroectodermal lineage causes specific depletion of cerebellar microglia (30). At present, it is not known whether CSF1-dependent and IL-34–dependent microglia differ in terms of phenotype and function.

TGF-β is another critical factor for microglial homeostasis. In the brain, TGF-β is produced by microglia, astrocytes, and neurons (31–33), whereas TGF-βR1 and TGF-βR2 are chiefly expressed in microglia (32–34). TGF-β knockout mice exhibit a deficit of microglia development (34). Likewise, ablation of TGF-βR2 in microglia using either the Sall1^CreER or the Cx3cr1^CreER conditional deletion systems disrupts the homeostatic morphology and phenotype of microglia (35–37). Mice with TGF-βR2–deficient microglia were also shown to develop white matter degeneration and a progressive limb paralysis during adulthood (36). Similarly, mice deficient for Lrrc33, a critical molecule for TGF-β signaling in macrophages, develop microglia alterations, progressive paralysis, and die prematurely (38, 39). At present, which component(s) of the SMAD transcriptional machinery are crucial for TGF-β signaling in microglia is not clear.

Several independent studies have shown that blood monocytes do not contribute to microglial turnover under homeostasis (3, 40, 41). However, monocytes can infiltrate the CNS and generate monocyte-derived macrophages in the presence of BBB damage (42–44). At steady-state, microglia self-renew through a slow but constant process of apoptosis and cell division in a stochastic manner (45–48). By contrast, microglia undergo clonal expansion under pathological conditions (47, 48). Microglia are rather long-lived cells, with a turnover rate of a few months that varies depending on the brain region they occupy. For example, the median lifespan of mouse cortical microglia is ∼15 mo (48). One study attempted to determine microglial turnover in the human brain and, although limited to only two subjects, found that human microglia proliferate at a very slow rate, with 0.08% of cells being replaced per day. Using a mathematical model, the average lifespan of human microglia was estimated to be ∼4.2 y, although some cells may be older than 20 y. This means that the entire microglial population is probably renewed multiple times during a human life (49).

Depletion of microglia, either by injection of diphtheria toxin into mice expressing diphtheria toxin receptor in macrophages or via chronic administration of CSF1R inhibitors, has provided insights into the dynamics of microglia turnover. Following depletion, microglia rapidly proliferate and repopulate the CNS in a few days (50–52). Determining the origin of repopulating microglia has been an active field of research for the past few years. Multiple studies determined that virtually all repopulating microglia originate from a few microglial cells that survive the depletion period, whereas blood monocytes do not contribute to repopulation, at least in the absence of BBB disruption because of irradiation or myeloablative chemotherapy (35, 53, 54). At present, the factors promoting microglia repopulation are obscure. However, it has been shown that deletion of either IL-1R1 or IKKβ (upstream kinase of the NF-κB cascade) in microglia significantly delays the repopulation (52, 54). In stark contrast to these studies, a recent work using the Cx3cr1^CreER26-DTR mice concluded that Ly-6C^hi monocytes do engraft the brain parenchyma in the absence of head irradiation. Monocyte-derived microglia were identified as F4/80^hiClec12a⁺, whereas resident microglia were F4/80^loClec12a⁻ (55). Similarly, another group used Cx3cr1^CreERCsf1r^flox/flox mice, which yielded partial depletion of microglia accompanied by monocyte recruitment into the CNS (56). It was also shown that irradiation-free transplantation of wild-type bone marrow into CSF1R-deficient pups (devoid of microglia) generates a massive invasion of donor-derived monocytes into the host CNS (57). Moreover, these studies have shown that resident and monocyte-derived microglia encompass phenotypically distinct populations. For example, unlike resident microglia, their monocyte-derived counterparts do not express the transcriptional regulator Sall1 (35, 55, 57, 58). Assay for transposase-accessible chromatin sequencing (ATACseq) revealed that certain loci that are typically open in resident microglia [such as Zfp691 (59, 60)] are transcriptionally inaccessible in monocyte-derived microglia (58). Conversely, monocyte-derived microglia express high levels of various immune-related genes (such as MHC class II [MHC-II] chains, Lyz2, Clec12a, Ms4a7, Apoe, Cybb) that are typically silenced in resident microglia at steady-state (55, 57, 58). Together, these data indicate that microglia and monocytes may contend for the colonization of empty niches in the brain. Under naive conditions, microglia greatly outcompete monocytes, perhaps by being more suited to the brain environment or simply because their homing into the CNS occurs during primitive hematopoiesis, long before adult hematopoiesis is established in bone marrow. Nevertheless, under conditions of microglia deficiency, monocytes may gain access to the CNS and differentiate into macrophages that partially resemble microglia (61).

Local environmental cues are critical for the final maturation and specialization of tissue-resident macrophages, including microglia (7, 62–66). Indeed, acutely isolated microglia lost their molecular identity within a few hours upon exposure to cell culture conditions (60, 67). Conversely, cultured microglia transplanted back into the mouse brain reacquired their original phenotype in about 2 wk (57, 67). Similarly, transplantation of induced pluripotent stem cells (iPSCs)-derived macrophages into the postnatal mouse brain generated microglia-like cells fully integrated within the host tissue (68, 69). Overall, the brain environment is critical for the maturation and maintenance of the microglial phenotype. Nevertheless, the failure of monocytes to acquire a complete microglial signature suggests that origin from the YS or bone marrow may imprint different repertoires of poised enhancers. For example, it has been shown that both miRNAs (70) and HDAC enzymes (71) critically shape microglia development during the embryonic stage but not after birth. Altogether, we can hypothesize that YS ontogeny dictates the epigenetic landscape in microglial progenitors, but the transcriptional signature is finally instructed within the CNS environment (57).

Bulk RNA-sequencing (RNAseq) studies identified a number microglia-specific genes, such as Crybb1, Fcrls, Gpr34, Gpr84, Hexb, Olfml3, P2ry12, P2yr13, Rnase4, Sall1, Siglech, Slc2a5, and Tmem119 (16, 32, 34, 59, 72). Nevertheless, caution should be used because evolving technologies for multidimensional analysis and deep sequencing always reveal previously unappreciated subpopulations. For example, Siglech is also a marker for plasmacytoid dendritic cells (DCs) (73). Although plasmacytoid DCs represent a minor population of meningeal DCs under homeostasis, encompassing ∼1.5% of the total MHC-II⁺ cells of the brain (74, 75), their number expands significantly during disease (76). Moreover, recent studies showed that Fcrls is broadly expressed in multiple brain macrophage subsets (76, 77), whereas Sall1 is apparently expressed in a population of macrophages within the apical choroid plexus epithelium (77).

More recently, single-cell RNAseq enabled more in-depth characterization of the transcriptional landscape in microglia at different stages of development. Embryonic microglia are enriched for various lysosomal enzymes, as well as Apoe and Ms4a7 (78, 79). Early postnatal microglia abundantly expressed Igf1, Spp1, Gpnmb, and Clec7a. Interestingly, secreted phosphoprotein 1 (Spp1)⁺Gpnmb⁺Clec7a⁺ postnatal microglia were primarily found within heavily myelinated regions such as the corpus callosum and cerebellum (78, 80). Of note, these studies consistently found a cluster of microglia particularly enriched for immediate early genes such as Fos, Jun, and Egr1 (77–80). It was, however, acknowledged that this cluster may have been artificially generated because of microglia activation during sample preparation (77, 80, 81).

One month after birth, mouse microglia are phenotypically mature with transcriptomes prominently enriched for homeostatic genes such as Tmem119, P2ry12, Slc2a5, Selplg, Cst3, Sparc, Tgfbr, Malat1, and others (78–80). Nonetheless, a region-dependent heterogeneity of microglia can be appreciated. For example, one study showed that microglia may vary phenotypically, depending on their topological distribution within the CNS (82). In particular, cerebellar microglia appeared skewed toward a more “immune alerted” and “metabolically demanding” phenotype, possibly because of the higher content of white matter as compared with other brain regions. By contrast, microglia in the cortex and striatum appeared in a more “quiescent” state, whereas hippocampal microglia had an intermediate phenotype. Another report identified transcriptomic heterogeneity in microglia selectively isolated from the cortex, nucleus accumbens, ventral tegmental area, and substantia nigra (83). Ingenuity pathway analysis identified the highest variability in pathways for vesicle release, mitochondrial function, cell metabolism, oxidative stress, lysosomal activity, and transport of metal ions. More recently, distinct patterns of gene-expression and epigenetic signatures were identified in cerebellar and striatal microglia. In particular, cerebellar microglia were highly enriched for genes linked to phagocytosis of apoptotic cells, whereas striatal microglia were more enriched for genes involved in immunological surveillance (84).

Last, molecular heterogeneity was recently described in human microglia. A mass cytometry study identified multiple region-specific subsets of microglia from postmortem brains (85). Consistently, single-cell RNAseq of microglia from healthy human brains formed multiple clusters with various enrichment for Tmem119, Cx3cr1, P2ry12, Slc2a5, Cst3, Ccl2, and Ccl4 (79). Understanding the functional implications of such microglial phenotypes will be an important challenge for the years to come.

Broad changes in the transcriptomic profile of microglia have been found in mouse models of amyloid pathology (86–88), Tau pathology (89), and experimental autoimmune encephalomyelitis (EAE) (74, 76, 88). Altogether, pathological conditions cause downregulation of the microglial homeostatic genes (including Tmem119, P2ry12, Selplg, Cx3cr1, Tgfbr1, and Sall1), whereas other genes are upregulated. For example, Trem2, Tyrobp, and Apoe were consistently found overexpressed in microglia in different neurodegeneration mouse models (86–92). Moreover, microglia exhibited Apoe upregulation during EAE (76, 88) and in the cuprizone model of toxic demyelination (79). The exact functions of TREM2 and APOE during brain diseases are under investigation, and this topic has been addressed elsewhere (93–96). Interestingly, independent studies identified a conserved molecular signature of microglia in models of amyloid pathology and neurodegeneration (86–88). This specific microglial phenotype has been termed “disease-associated microglia” (DAM) (86).

In mice, the DAM signature is characterized by higher expression of genes involved in lysosomal functions (Cst7, Ctsb/d), phagocytosis (Axl), Ag presentation (H2-Aa, H2-Ab), lipid transport (Lpl, Apoe), matrix remodeling (Spp1, Gpnmb), complement binding (CD11c), antimicrobial activity (Lyz2, Dectin-1), immune modulation (leukocyte Ig-like receptor B4 [Lilrb4]), and cell survival (Csf1, Igf1) (74, 77, 86–88) (Fig. 2).

Cystatin F (Cst7) is a lysosomal cysteine protease inhibitor, which targets the lysosomal enzyme cathepsin C (97). Inhibition of Cst7 expression correlates with a reduced amyloid pathologic condition (98), indicating that Cst7 may reduce the capacity of microglia to degrade Aβ. Alongside, DAM signature is also characterized by higher expression of some cathepsins, especially Ctsb and Ctsd (86). Cathepsins are cysteine proteases important for lysosomal degradation of aggregated proteins (99). Additionally, cathepsins can be secreted and may play a role in cell migration (100).

Axl and Mertk belong to a family of TAM tyrosine kinases receptors mainly involved in the phagocytosis of dead cells (101). Axl is upregulated in microglia during neurodegeneration and neuroinflammatory diseases (86–88, 102), whereas Mertk is downregulated (74, 86, 88). Microglia lacking Axl and Mertk exhibit deficient phagocytosis of apoptotic cells and display reduced migration toward laser-induced injury (102, 103). In the EAE model, Axl-deficient mice show more severe pathology and fewer macrophages infiltrating the spinal cord (104). Whether Axl or Mertk are directly involved in the uptake of protein aggregates (such as Aβ and α-synuclein) is unknown.

Spp1(osteopontin) is one of the most upregulated genes in the DAM signature (77, 86–88), and significantly increased levels of Spp1 were found in the CSF of Alzheimer disease (AD) and frontotemporal dementia patients (105, 106). In the periphery, Spp1 is highly expressed by both osteoblasts and osteoclasts and is important for the bone mineralization (107, 108). Spp1 is also secreted by different leukocytes, including Th1 cells, macrophages, and DCs (109). Spp1 induces IL-12 production in DCs, thus promoting type 1 immunity (110, 111). Moreover, Spp1 was shown to improve survival of autoreactive T cells in EAE (112). However, it was also suggested that Spp1 regulates inflammatory reactions (113), for example via inhibition of NO production in macrophages (114). Spp1 is also known to bind CD44 (110, 115), which in the brain is primarily expressed by astrocytes (32, 33, 116), whereas the expression in microglia is negligible (117). Spp1 and CD44 may then form a communication axis between microglia and astrocytes during neurodegeneration. Additionally, secreted Spp1 represents a substrate for the activation of metalloproteinases (118), suggesting that it may play some role in microglia migration toward the injury site. Further studies on conditional Spp1-deficient mice are required to better understand the function of this protein in microglia. Similar to Spp1, osteoactivin (Gpnmb) represents another possible ligand for CD44 (119). It was suggested that Gpnmb may dampen astrocyte activation via CD44 signaling (120).

Clec7a (Dectin-1) is a C-type lectin serving as pattern recognition receptor against fungi and bacteria (121). Clec7a contains an ITAM motif promoting a Syk-dependent signaling, which elicits an antimicrobial response in macrophages (122, 123). Albeit not expressed under homeostasis, Clec7a⁺ microglia were found in proximity of amyloid plaques (86, 88) as well as in mice with microglia-specific BRAF mutation, which causes microgliosis and late-onset neurodegeneration (124). In vitro, microglial metabolic activity is boosted by zymosan, which is a known ligand for Clec7a (125). Possibly, Clec7a may help mount the immune-activation state of plaque-associated microglia. Moreover, Aβ is known to induce microglia activation via various pattern recognition receptors (126). It remains unclear whether Clec7a is involved in a similar mechanism.

CD11c (Itgax), a prototypical DC marker, has repeatedly been found upregulated in activated microglia and represents one of the most consistent DAM signature genes (74, 77, 86, 88). CD11c and CD18 form the complement receptor 4, which is important for the engulfment and fragmentation of complement-opsonized particles (127). It is then tempting to speculate that CD11c in plaque-associated microglia may help recognize or phagocytose Aβ aggregates. A conditional knockout mouse model is needed to better investigate the role of CD11c in microglia under pathology.

Lilrb4 belongs to a family of ITIM-bearing inhibitory receptors widely expressed in different leukocytes (128). At present, the ligand as well as the exact function of mouse Lilrb4 is unknown. A recent study showed that conditional ablation of Lilrb4 exacerbates steatosis and systemic inflammation in mice under high fat diet (129), suggesting that this receptor may play important immunomodulatory functions. Whether Lilrb4 could modulate the microglia activation state remains hypothetical.

Lastly, the DAM signature shows an enrichment for Csf1 and Igf1. Interestingly, both genes are highly expressed in microglia at the early stages of brain development (15, 80, 130, 131). Csf1 is important for microglial survival and proliferation (51, 132), whereas microglia-derived Igf1 was shown to support the survival of newborn neurons (130). Csf1 could act on microglia in an autocrine/paracrine manner, thus promoting microglia proliferation/survival within the plaque-surrounding environment. RNAseq data indicate that expression of Igf1r in microglia is negligible but detectable in neurons (32, 33). Microglia-derived Igf1 could then provide neurotrophic support to the neighboring neurons, thus protecting against Aβ cytotoxicity.

Our understanding of the molecular properties of microglia relies chiefly on mouse models. At present, only a few pioneering works have delineated the transcriptomic profiles of microglia from healthy subjects as well as from patients with AD and multiple sclerosis (MS) (60, 79, 116, 133). A seminal study showed that the transcriptomes of both mouse and human microglia are characterized by a dominant PU.1-dependent signature, and ∼50% of the microglia-specific genes (such as Cx3cr1, Tmem119, P2ry12, Trem2, and Sall1) are similarly expressed in both species. Using a cutoff of 10-fold, 2.5% of the transcripts were highly enriched in human microglia (such as C3, SPP1, and APOE), and 1.9% appeared specific for mouse microglia (such as Hexb, Sparc, and Sall3). Overall, these data indicate a substantial overlap between mouse and human microglia at the molecular level (60). Similar findings were reported in a following study that, however, highlighted increasing molecular disparity between mouse and human microglia during aging (134). A mass cytometry study showed that human microglia express high levels of TMEM119 and P2RY12, which are absent in blood myeloid cells. Human microglia are also positive for EMR1 (F4/80), TREM2, CX3CR1, CD64, and CD115, whereas expression of CD45, CD44, CCR2, CD206, and CD163 is either low or negligible. This repertoire of surface markers closely resembles that of mouse microglia. However, unlike their mouse counterparts in the steady-state, human microglia express CD11c, MHC-II, and relatively low levels of CD11b (85).

In AD patients, microglia upregulate CD74, HLA-DR, APOE, TREM2, C1Q, and CD14. Interestingly, transcriptomic changes in microglia seemed to correlate with the severity of both amyloid and Tau pathologies (133). Given the difficulties of working with human brain samples, it has been suggested that iPSCs may be a powerful tool for modeling human microglia during brain diseases (68, 69, 135, 136). For example, a recent study showed that following transplantation into the brains of mice with amyloid pathology, iPSC-derived human microglia efficiently migrated toward amyloid plaques and, like their murine counterparts, upregulated APOE, HLA-DR, LGALS3, MS4A7, ITGAX, and TREM2. Of note, some DAM signature genes, such as TYROBP, CST7, CLEC7A, and CSF1, were not significantly altered in iPSC-derived human microglia (137). This suggests that mouse and human microglia mount similar immunological responses against amyloid pathology in vivo; however, certain pathways may not be conserved between species. Our own data support this view, as we detected a prominent IRF8-dependent signature in human microglia from AD patients but not in mouse models (138). Further studies are required to better understand similarities and differences in the DAM signatures of human and mouse microglia.

Microglia are not the only brain-resident macrophages; indeed, populations of extraparenchymal macrophage patrol the brain–blood interface in both mice and humans. Perivascular macrophages (pvMPs) are primarily found in the perivascular Virchow–Robin spaces of the cortical blood vessels. Meningeal macrophages (mMPs) are located within the meningeal membranes, either on the pia mater or within the dura. Last, choroid plexus macrophages (cpMPs) lie beneath the epithelial cell layer of the choroid plexus (139). These extraparenchymal brain macrophages are collectively referred to as border-associated macrophages (BAMs) (74). Although microglia and BAMs share the expression of several phenotypic markers (including Iba1, CD11b, CX3CR1, CD64, Mertk, CD115, and others), transcriptome studies identified a repertoire of molecules that are specifically expressed by each population (Fig. 3). Like microglia, BAMs are dependent on CSF1R signaling for their survival (74, 77). However, it is still unknown whether anatomically distinct BAM subsets are differentially dependent on CSF1 and IL-34. Interestingly, genetic deletion of the superenhancer fms-intronic regulatory element (FIRE), which is critical for the expression of CSF1R in YS-derived macrophages, causes complete depletion of microglia but leaves pvMPs and mMPs seemingly unaffected. This may suggest that transcriptional and epigenetic mechanisms differentially regulate CSF1R in microglia and extraparenchymal macrophages (140).

It has long been thought that BAMs are derived from and constantly replaced by bone marrow–derived monocytes. However, recent fate-mapping studies revealed that, like microglia, BAMs are generated by YS-derived progenitors. Embryonic-derived pvMPs and mMPs are long-lived cells, whereas cpMPs are rapidly replaced by bone marrow–derived monocytes soon after birth (141). Of note, a recent paper described a population of BAMs localized on the apical choroid plexus epithelium (previously identified as epiplexus Kolmer cells) that share key features of microglia. These cells are embryonic derived, express Sall1, and can self-renew with no obvious input from the periphery. By contrast, stromal cpMPs are negative for Sall1 and undergo constant monocyte-mediated turnover (77). At present, it is unclear whether YS-derived myeloid progenitors are already committed to become either microglia or BAMs during the embryonic development or whether the two differentiation pathways are locally instructed by environmental stimuli. The first case predicts the existence of a common myeloid progenitor (possibly downstream of the EMP stage) that generates two separate lineages. Alternatively, we may hypothesize that the local environment of the blood–brain interface guides the differentiation of embryonic macrophages toward a BAM phenotype, whereas the parenchymal environment supports the differentiation into microglia (Fig. 4). Future studies will shed some light on this outstanding question.

The composition of BAMs during neuroinflammatory disease has recently been investigated by single-cell RNAseq and multidimensional immunophenotyping techniques. During EAE, for example, expression of several activation markers (MHC-II, CD74, CD44, Sca-1, and CD11c) was increased, whereas the BAM homeostatic markers Lyve-1, Fcrls, and CD206 were downregulated. Expression of other markers like Ms4a7, CD38, CD169, and CD86 remained stable during disease (74, 76). Meningeal myeloid cells have often been implicated in the reactivation of autoaggressive T cells during EAE (142–145). Nonetheless, experimental proof that BAMs license T cell entry into the CNS is lacking. A landmark publication used intravital two-photon imaging to illustrate the path of autoreactive T cells in a rat model of transfer EAE (146). Both MBP- and MOG-reactive T cells entered the CNS by crawling along the leptomeninges, where they randomly encountered mMPs. Given that the leptomeninges represent a main gateway for infiltrating T cells, mMPs probably constitute the first line of APCs at the brain–blood interface. TCR engagement by meningeal APCs induced nuclear translocation of NFAT in Ag-competent T cells, thereby eliciting their encephalitogenic potential (146). Intrathecal administration of blocking Ab for LFA1/VLA4 integrins augmented the number of MBP-reactive T cells in the CSF but dramatically reduced their infiltration into the underlying parenchyma (146). Consistent with this, in a rat model of gray matter inflammation, the number of synuclein-reactive T cells was diminished in both the parenchyma and meninges after intrathecal injection of anti–MHC-II Ab. In addition, this treatment significantly attenuated expression of inflammatory cytokines (IFN-γ and IL-17) as well as the EAE clinical score (147). Altogether, these findings indicate that the meningeal myeloid compartment represents a critical checkpoint for the restimulation of autoreactive T cells before their invasion of the CNS parenchyma. To more precisely interrogate the Ag-presentation capacity of microglia and BAMs during EAE, three independent groups studied the effect of MHC-II deletion in microglia using either the Cx3cr1^CreER or the Sall1^CreER inducible systems. EAE pathologic condition was unaffected in mice with MHC-II–deficient microglia in all three studies (75, 76, 148). By contrast, deletion of MHC-II using the CD11c^Cre line (which targets both DCs and activated macrophages) completely abolished the onset of paralysis, CNS infiltration, and demyelination (76). Additional studies are now required to clarify whether meningeal DCs are uniquely competent APCs in the brain or whether other BAMs also play a role in this context. Nevertheless, these data strongly indicate that microglia are probably dispensable for the reactivation of T cells during EAE.

During neuroinflammation, microglia might significantly contribute to tissue damage by releasing inflammatory cytokines. A mass cytometry study of EAE identified seven distinct microglial populations expressing variable levels of proinflammatory (mostly TNF-α, IL-6, and GM-CSF) and anti-inflammatory (TGF-β and IL-10) cytokines (149). This may indicate that microglia both promote and help resolve neuroinflammation during EAE. Furthermore, a recent study based on two-photon intravital microscopy during EAE suggested that CNS phagocytes shift from a proinflammatory to a wound-healing phenotype, depending on the lesion context (150). A seminal study on Cx3cr1^CreERTak1^flox/flox mice showed that deletion of the NF-κB activator Tak1 in microglia remarkably ameliorated EAE pathology in mice (41), which implies that microglia can be gravely detrimental in neuroinflammatory conditions. However, like microglia, meningeal and pvMPs are CX3CR1⁺ and undergo very limited turnover (141). Therefore, the observed protective effect could partially stem from Tak1 deletion in BAMs rather than microglia. In contrast, deletion of NF-κB negative regulators (such as AHR or A20) in microglia was shown to exacerbate neuroinflammation and EAE pathology (151, 152). Conversely, the ubiquitin-specific peptidase 18 (USP18) was shown to act as a negative regulator of the Stat1 pathway in white matter microglia, thus dampening the type I IFN response (153). Thus, microglia lacking USP18 constitutively upregulate IFN-dependent genes, resulting in white matter disease and behavioral defects (154).

Besides their proinflammatory function at the peak of the disease, microglia may play a critical role in the resolution of inflammation and tissue repair. Microglia were repeatedly shown to be important for the clearance of myelin debris and ensuing remyelination both in EAE (155) and toxic-induced demyelination models (156–160). This suggests that microglia can support the healing of the white matter after demyelination. A recent study showed that a population of CD11c⁺Clec7a⁺Gpnmb⁺ microglia promote remyelination via secretion of Igf1 during EAE (131). Additionally, microglia were shown to upregulate PD-L1 in the EAE model (74), whereas a microglial subset expressing galectin-1 was found in postmortem brain samples from MS patients (79). Both of these genes critically modulate the activation of CD8⁺ T cells (161–164). It is tempting to speculate that microglia could help dampen T cell–mediated cytotoxic response during MS. Of note, microglia depletion approaches have generated conflicting outcomes in the context of EAE, resulting in either beneficial, detrimental, or no effects (165–167). We would hypothesize that nonselective depletion of both microglia and BAMs may produce confounding results.

Altogether, microglia may promote tissue damage during neuroinflammation via cytokine and reactive oxygen species production. For example, recent studies suggested a pathogenic role of the inflammasome machinery in microglia during amyloid and Tau pathology (168, 169). However, microglia are also critical for efficient scavenging of cellular debris and tissue regeneration. Environmental cues that selectively promote these functions are currently under investigation.

Brain macrophages encompass multiple populations characterized by different anatomical distribution, phenotype, ontogeny/turnover, and very likely, different functions. Growing evidence suggests that molecular signature of microglia and BAMs is instructed by a combination of local environment and ontogeny. However, whether microglia and BAMs arise from a unique YS-derived progenitor or develop through independent pathways is still unknown (170). Additionally, we just started to scratch the surface of the transcriptomic changes in brain macrophages during disease. For example, a number of independent works provided a list of candidate genes that identifies the DAM signature of plaque-associated microglia during amyloid pathology. Nevertheless, the exact function of these genes remains to be determined. This information may help us harness specific DAM genes to either reduce the amyloid burden or improve the viability of neighboring brain cells in AD.

We thank Dr. Susan Gilfillan for helpful suggestions during the preparation of the manuscript.