Conventional DCs sample and present myelin antigens in the healthy CNS and allow parenchymal T cell entry to initiate neuroinflammation

See allHide authors and affiliations

Science Immunology  25 Jan 2019:
Vol. 4, Issue 31, eaau8380
DOI: 10.1126/sciimmunol.aau8380

Licensing myelin-reactive T cells

Multiple distinct populations of potential antigen-presenting cells (APCs) are interspersed among the different anatomical components of the brain, including microglia, B lymphocytes, macrophages, and dendritic cells (DCs). Mundt et al. investigated which steady-state APC types are responsible for displaying peptide fragments of myelin proteins to pathogenic CD4+ T cells with the capacity to initiate neuroinflammatory disorders such as human multiple sclerosis. Adoptive transfer of myelin-reactive CD4+ T cells to mice with conditional deletion of MHC class II molecules in specific brain APC subsets identified conventional DCs as the essential APCs enabling initiation of T cell–mediated immunopathology. The results of this study will assist in the precision targeting of immunotherapies aimed at restraining rogue T cells responsible for human neuroinflammatory diseases.


The central nervous system (CNS) is under close surveillance by immune cells, which mediate tissue homeostasis, protection, and repair. Conversely, in neuroinflammation, dysregulated leukocyte invasion into the CNS leads to immunopathology and neurological disability. To invade the brain parenchyma, autoimmune encephalitogenic T helper (TH) cells must encounter their cognate antigens (Ags) presented via local Ag-presenting cells (APCs). The precise identity of the APC that samples, processes, and presents CNS-derived Ags to autoaggressive T cells is unknown. Here, we used a combination of high-dimensional single-cell mapping and conditional MHC class II ablation across all CNS APCs to systematically interrogate each population for its ability to reactivate encephalitogenic TH cells in vivo. We found a population of conventional dendritic cells, but not border-associated macrophages or microglia, to be essential for licensing T cells to initiate neuroinflammation.


Multidirectional interactions between the central nervous system (CNS) and the immune system have a profound impact on brain protection and pathology. There is evidence that not only brain development and homeostasis but also social behavior, psychiatric disorders, and neurodegenerative diseases such as Alzheimer’s disease and Parkinson’s disease are influenced by innate and adaptive immunity (1). Multiple sclerosis (MS) is a prototypical paradigm for pathological CNS-immune interactions, because CNS-invading autoaggressive T helper (TH) cells induce immunopathology, leading to demyelination, axonal loss, and brain atrophy (2). Even fully primed and activated encephalitogenic TH cells must encounter their cognate antigen (Ag) in the CNS to cross the basal lamina of the blood-brain barrier (BBB) and to infiltrate the parenchyma (3). This local reconfirmation of the correct Ag then also triggers an inflammatory cascade and the recruitment of monocytes, which, in turn, mediate CNS tissue damage (3). The reactivation of myelin-reactive TH cells outside but proximal to the parenchyma requires a CNS-resident population of major histocompatibility complex (MHC) class II (MHCII)–expressing Ag-presenting cells (APCs), which have sampled myelin protein from oligodendrocytes and present its peptides within the homeostatic CNS.

Within the CNS parenchyma, microglia are the only resident leukocytes and therefore the most likely candidate to directly interact with invading T cells (4). Recently, high-dimensional cytometry and algorithm-guided analyses revealed an abundant and complex immune cell landscape of CNS-blood and CNS–cerebrospinal fluid interfaces, which includes border-associated macrophages (BAMs), monocyte-derived cells (MdCs), dendritic cells (DCs), lymphocytes, and even granulocytes (58).

The prototypical tool to study T cell–mediated neuroinflammation in vivo is experimental autoimmune encephalomyelitis (EAE), which not only serves as the preclinical model for MS but also permits the investigation of T cell–APC interactions in general. Most EAE-based studies that identify CNS APCs have used bone marrow chimeric rodents, which—largely due to the irradiation-induced cytokine storm—suffer from the infiltration of the CNS by various immune cells and, therefore, do not allow the study of T cell–Ag encounters within the steady-state CNS (9). Likewise, active immunization with complete Freund’s adjuvant (CFA) also changes the CNS leukocyte landscape independent from the presence of auto-Ag (10). Complicating matters is the difficulty of distinguishing between CNS-resident and CNS-invading leukocytes because of overlapping phenotypes. We had previously presented data suggesting that CNS-associated DCs are the most potent APCs of the CNS (7). However, CD11c, which we had used to restrict MHCII expression to DCs, has been shown to not be specific to DCs, because microglia, BAMs, Ly6Chi, and Ly6Clow monocytes—which are all present in the steady-state CNS—can also express this marker (5, 1113). Moreover, there is some evidence for APC function in BAMs, microglia, and other CNS-resident cells (1416).

To systematically interrogate each potential CNS APC population for its capacity to effectively present myelin Ag to encephalitogenic TH cells, we took advantage of new in vivo genetic targeting and single-cell cytometry tools. We first performed mass cytometric [cytometry by time of flight (CyTOF)] analysis with subsequent high-dimensional data mining to identify and characterize all APCs of the steady-state brain. Conditional deletion of MHCII across CNS APCs in combination with adoptive transfer (AT) EAE allowed us to specifically investigate T cell reactivation in the CNS, independent of peripheral priming.


Multiple potential APCs are found in the steady-state CNS

To determine which cells of the CNS have the potential to present Ag, we scanned all CNS-associated leukocytes for MHCII expression in the steady-state brain. To be able to fully capture the complexity of CNS immune cells, we used our recently described workflow for the high-dimensional single-cell analysis by mass cytometry and computational data mining (5). Live, single cells were isolated from murine brains (C57BL/6), stained with metal-tagged antibodies, and acquired with a CyTOF analyzer. After preprocessing and manual gating of CD45+ MHCII+ cells (Fig. 1A), the data were visualized on a t-distributed stochastic neighbor embedding (t-SNE) map (17), and the cells were grouped into populations by Flow-Self Organizing Map (SOM)–guided clustering for an unbiased identification and overview of the major MHCII-expressing leukocyte populations in the CNS (5, 18, 19). We identified eight distinct APC populations by their specific expression profile (Fig. 1, B and C, and fig. S1A): BAMs (42%), CD172a+ conventional type 2 DCs (cDC2s) (25%), Ly6Cint MdCs (7%), mature B cells (10%), a microglia subpopulation (6%), CD24hi immature B cells (6%), CD11blo cDC1s (2%), and plasmacytoid DCs (pDCs) (1.5%) (Fig. 1D). MHCII expression on microglia, pDCs, and immature B cells was low compared with moderate MHCII levels on MdCs and B cells and high expression on BAMs and cDCs (Fig. 1E). CD86 expression was low on immature B cells, B cells, and microglia, was moderately expressed in BAMs, and was highest in pDCs and cDCs, whereas MdCs displayed an exceptionally activated phenotype with high co-inhibitory programmed death-ligand 1 (PD-L1) expression (Fig. 1E).

Fig. 1 Identification of MHCII+ cells in the steady-state CNS.

(A) Immune cell populations from the whole steady-state brain (i.e., parenchyma, parenchymal vessels, intact choroid plexus, and the adhering leptomeninges) including the dura mater were analyzed by CyTOF. After gating on live singlets, MHCII+ CD45+ cells were exported as an FCS file using the FlowJo software. (B) The MHCII+ fraction of CNS immune cells was visualized using t-SNE and clustered using the FlowSOM algorithm in R (6907 total cells: max iterations = 5000, perplexity = 50, θ = 0.5). (C) Median relative expression of all panel markers. (D) Frequency of CD45+ (left graph) and MHCII+ cells (right graph) of FlowSOM clusters. (E) FlowSOM cell clusters were analyzed for MHCII, CD86, and PD-L1 expression intensity (after percentile normalization) and plotted as histograms for each cluster using R. Shown are pooled data from n = 3 mice.

Regarding the localization of MHCII-expressing cells across the CNS, we found immature B cells, pDCs, and MdCs to be abundant in the dura mater of the naïve CNS, whereas other brain compartments were almost devoid of these cell types (fig. S1B). In contrast to the inner layer of leptomeninges, the dura mater is not involved in lymphocyte trafficking into the CNS (20, 21). We therefore excluded the dura mater from further experiments and focused our analysis on cell types, which do not exclusively reside in this compartment.

Because of their strategic location at the interface of systemic circulation and the CNS parenchyma, brain endothelial cells were also considered as potential APCs (14). However, we failed to detect MHCII expression on CD31+ cells in the steady-state brain (fig. S1, C and D) or in active neuroinflammation (fig. S1E). We thus also excluded endothelial cells from further detailed interrogation.

Microglia are dispensable for T cell reactivation within the CNS

Microglia globally up-regulate MHCII during neuroinflammation, revealing their potential as APCs (fig. S2A) (5). In the steady state, however, the bulk of microglia (97 to 100%) did not express MHCII (fig. S2, A and B). The small subset of MHCII+ microglia occasionally detected in the brain of naïve mice expressed only low MHCII (Fig. 1, B, C, and E, and fig. S2B) (5). We recently described Sall1 as a microglial signature gene allowing microglia-specific manipulation (22). Using Sall1GFP reporter mice, we confirmed the identity of MHCII+ microglia, which we had identified by FlowSOM clustering (Fig. 1B and fig. S2B). Next, we crossed Sall1CreERT2 mice to Iabfl mice carrying a conditional loxP-flanked Iab allele (23) (Fig. 2A) to specifically delete MHCII in microglia (Fig. 2B) and to investigate the role of the small steady-state MHCII+ microglial subset on myelin Ag presentation to CNS-infiltrating T cells. Because of the small and varying numbers of MHCII-expressing microglia, the targeting efficiency of this system could not be determined in the naïve brain. MHCII expression of non-microglial APCs remained unchanged in Sall1CreERT2/+ Iabfl/fl mice (Fig. 2C).

Fig. 2 Microglial MHCII is dispensable for AT EAE.

(A) Sall1CreERT2 mice were crossed to Iabfl mice carrying floxed MHCII alleles. (B) Graphical abstract to illustrate the tamoxifen treatment strategy of Sall1Cre/+ Iabfl/fl mice. Cre+ animals and Cre littermates were administered tamoxifen via oral gavage (three times on alternate days) to induce Cre recombinase expression in microglia. (C) MHCII targeting profile of whole steady-state brain APCs in Sall1CreERT2/+ Iabfl/fl and Sall1+/+ Iabfl/fl mice 1 week after tamoxifen treatment (for gating strategy, see fig. S2, C and D). The relative median fluorescence intensity (MFI) has been calculated by normalizing absolute MFIs of each cell population to the MFI of cDC2s (which do express the highest MHCII levels). The MFI of cDC2s of one mouse was set to 100%. Data are representative of one of two independent experiments (n > 7 mice per group). (D) EAE was induced by AT of encephalitogenic 2D2 cells into recipient mice. Data show individual maximal EAE score and number of mice with clinical EAE symptoms. (E to G) At the peak of disease, the CNS (whole brain and spinal cord pooled) was analyzed for inflammatory infiltrates (E) and MHCII targeting of APCs (F and G) by flow cytometry. (E) Infiltrating monocytes (Ly6C+, Ly6G, CD11b+), neutrophils (Ly6Cint, Ly6G+, CD11b+), and CD4+ T cells (CD4+, CD11b, CD11c) were manually gated in FlowJo; absolute numbers (log10) per brain ± SEM are shown. (F) MHCII targeting profile (log10 MFI MHCII) in microglia and (G) other CNS-infiltrating and CNS-resident APCs at peak EAE (relative MFI) (manually gated in FlowJo without using MHCII for the gating). (D to G) Data are pooled from three independent experiments (n > 9 mice per group).

We adoptively transferred lymphocytes from actively immunized T cell receptor (TCR) transgenic 2D2 mice (24) into tamoxifen-treated Sall1CreERT2/+ Iabfl/fl and littermate controls (Fig. 2B). The TCR in 2D2 mice recognizes the myelin oligodendrocyte glycoprotein (MOG)35–55 peptide in the context of MHCII. We developed an AT EAE protocol omitting the irradiation of recipient mice (25), thereby preserving steady-state conditions of the CNS and the integrity of the BBB. Both Sall1CreERT2/+ Iabfl/fl mice and Sall1+/+ Iabfl/fl littermates developed EAE with similar kinetics, clinical scores (Fig. 2D), and inflammatory CNS infiltration (Fig. 2E). This is in line with a recent report by Wolf and colleagues (26), which, using a different targeting approach, also showed MHCII in microglia to have no function in T cell entry during EAE or demyelination. Activation-induced microglial MHCII expression was completely aborted in Sall1CreERT2/+ Iabfl/fl mice (Fig. 2F), whereas other cell types retained MHCII (Fig. 2G), confirming the specificity of Sall1CreERT2 for parenchymal microglia.

The Iab locus is hypersensitive to Cre-mediated targeting

Having excluded microglia as vital APCs in neuroinflammation, we next aimed to target non-microglial CNS-associated APCs (DCs, BAMs, and B cells). First, to delete MHCII expression in BAMs, we crossed the myeloid cell–specific Lyz2Cre mice to the Iabfl strain (fig. S2E) (27). In the steady-state brain of Lyz2Cre/+ Iabfl/fl mice, MHCII expression not only on Lyve1 BAMs (containing the MHCII+ BAM subset) (5) but also on other APCs, including cDC1s, cDC2s, and even B cells, was strongly diminished when compared with littermate controls (fig. S2G; gating strategy: fig. S2, C and D), indicative of promiscuous gene targeting outside the target populations. AT of encephalitogenic T cells into these Lyz2Cre/+ Iabfl/fl mice expectedly revealed protection from EAE (fig. S2G). We next used ItgaxCre/+ Iabfl/fl mice to specifically delete MHCII expression in CD11c-expressing APCs (28), namely, DCs in the steady-state CNS. However, similar to what we observed with the Lyz2Cre system, MHCII was deleted across the entire APC compartment of the steady-state brain (fig. S3H). In addition, the ItgaxCreEGFP/+ Iabfl/fl strain (29), which has been generated by a knock-in strategy to deliver a more faithful targeting pattern, resulted in an overreaching of the targeting frequency (fig. S3I). The same problem arose in Zbtb46Cre/+ Iabfl/fl mice, where Cre expression is directed by the Zbtb46 gene, a transcription factor reported to be exclusively expressed in cDCs and endothelial cells (fig. S3J) (30). These findings demonstrate that constitutive Cre-mediated targeting of Iab is unreliable in that it leads to almost complete ablation of MHCII expression across numerous cell types.

Deleting MHCII in CNS cDCs, but not macrophages, confers protection to AT EAE

To overcome the limitations of the hypersensitive Iab locus, we opted to test an inducible targeting system. Taking advantage of the differences in cell longevity, the Cx3cr1CreERT2 strain permits the specific targeting of CX3CR1+ macrophages (including BAMs and microglia), Ly6Clow monocytes, and DCs (8, 31) by inducing recombination at different time points (Fig. 3, A and B). We used two tamoxifen treatment protocols to differentially target CX3CR1+ MHCII+ cells: an “early” and a “continuous” regimen.

Fig. 3 cDCs, but not BAMs or microglia, are required for the reactivation of 2D2 CD4+ T cells in the CNS.

(A) Cx3cr1CreERT mice were crossed to Iabfl mice. (B) Cx3cr1CreERT2/+ Iabfl/fl (early) mice and Cx3cr1+/+ Iabfl/fl littermates received a short, early tamoxifen treatment via oral gavage (three times on alternate days) 4 weeks before AT EAE or steady-state CNS analysis. Tamoxifen treatment of Cx3cr1CreERT2/+ Iabfl/fl (cont.) and Cx3cr1+/+ Iabfl/fl littermates was initiated 1 week before induction of AT EAE or steady-state brain analysis and continued until the end of the experiment. (C) MHCII targeting profile in APCs of naïve Cx3cr1+/+ Iabfl/fl (white circles) and Cx3cr1CreERT2/+ Iabfl/fl mice (early: gray circles; cont.: red circles) from the whole steady-state brain. Data represent the MFI (±SEM) of MHCII of each manually gated (for gating strategy, see fig. S2, C and D) population and are representative of one of more than three independent experiments (total of n > 12 mice per group). (D) Encephalitogenic 2D2 cells were adoptively transferred, and the clinical outcome was compared between Cx3cr1CreERT2/+ Iabfl/fl mice (early, gray circles; cont., red circles) and Cx3cr1+/+ Iabfl/fl littermates (white circles) over time. Data represent the individual maximal EAE score and the number of mice with clinical EAE (pooled from three independent experiments with n > 14 mice per group); P < 0.0001, χ2 analysis. (E) At the peak of disease (days 12 to 14), the CNS (whole brain and spinal cord pooled) was analyzed for inflammatory infiltrates by flow cytometry. Infiltrating monocytes (Ly6C+, Ly6G, CD11b+), neutrophils (Ly6Cint, Ly6G+, CD11b+), and CD4+ T cells (CD4+, CD11b, CD11c) were manually gated in FlowJo, and absolute numbers (log10) per brain ± SEM are shown. Data are representative of one of three independent experiments (total of n > 10 mice per group).

In the early treatment regimen, mice were exposed to tamoxifen 4 weeks before AT of encephalitogenic T cells [referred to as Cx3cr1CreERT2/+ Iabfl/fl (early)] (Fig. 3B) to target long-lived, self-maintaining CX3CR1+ cells, namely, microglia, and perivascular and meningeal BAMs in the CNS; short-lived DCs and monocytes are replaced over time and will not be targeted at the time point of AT (8, 31). Accordingly, MHCII expression was specifically reduced in BAMs (−89%) of the steady-state brain, whereas DCs, B cells, and other APCs retained MHCII expression (Fig. 3C; gating strategy: fig. S2, C and D). After AT of encephalitogenic T cells, Cx3cr1CreERT2/+ Iabfl/fl (early) mice and tamoxifen-treated littermates developed similar signs of EAE with comparable incidence, mean maximum score (Fig. 3D), and inflammatory CNS infiltration (Fig. 3E).

In the continuous treatment regimen, mice were exposed to tamoxifen 1 week before steady-state CNS analysis and treatment was continued during AT EAE [referred to as Cx3cr1CreERT2/+ Iabfl/fl (cont.)], to deplete MHCII expression across all CX3CR1+ cells, regardless of their ontogeny or lifespan. Accordingly, we observed that MHCII expression was abrogated not only in BAMs (−94%) but also in cDC1 cells (−65%) and cDC2 cells (−72%) (Fig. 3C; gating strategy: fig. S2, C and D). cDC1s and cDC2s display low-to-moderate CX3CR1 expression in the steady-state brain and are thus targeted with continuous tamoxifen treatment (fig. S3A). Continuous tamoxifen treatment resulted in markedly reduced clinical EAE symptoms in Cx3cr1CreERT2/+ Iabfl/fl (cont.) mice compared with tamoxifen-treated littermates (Fig. 3D), accompanied by reduced CNS infiltrates at the peak of EAE (Fig. 3E). Continuous tamoxifen treatment itself had no impact on the clinical outcome of AT EAE (fig. S3B). Microglia of Cx3cr1CreERT2/+ Iabfl/fl mice did not up-regulate MHCII expression during EAE, regardless of the treatment regimen, thus confirming the expected targeting of microglia with both strategies (fig. S3C). Together, differential targeting of CX3CR1-expressing cells in the CNS revealed that, in contrast to microglia and BAMs, cDCs are required for the reactivation of encephalitogenic T cells and the initiation of CNS infiltration and neuroinflammatory events in EAE.

Distinct subsets of DCs reside at CNS interfaces

Because of a lack of specific markers, traditional methods have thus far not allowed a thorough characterization of DCs in the steady-state CNS. We further characterized the CNS DCs by enriching for this rare cell population by fluorescence-activated cell sorting (FACS) from the steady-state brain and then categorized the cells using our mass cytometry workflow with a DC subset–focused antibody panel. Here, we confirmed the existence of three major populations: cDC1s, cDC2s, and pDCs. The vast majority of DCs were CD11b+ CD172a+ cDC2s (76%), as observed in other peripheral tissues (Fig. 4, A and B) (32). We also identified two distinct CD11blo cDC1 subpopulations, which differ with respect to XCR1, CD103, and CD205 expression. Last, Ly6C+, B220+, Siglec-H+ pDCs could be clustered into MHCIIint and MHCIIlow cells.

Fig. 4 CNS-associated cDC2s reside in the steady-state brain leptomeninges and reactivate 2D2 T cells in vitro.

(A) In-depth CyTOF analysis of DC subsets from pre-enriched whole-brain leukocytes (n = 20 naïve brains): DCs were identified and subset from the initial data, subjected to t-SNE dimensionality reduction (3509 total cells; max iterations = 750, perplexity = 200, θ = 0.5), and clustered into three main subsets using FlowSOM-guided clustering according to their marker expression. (B) Relative abundance of the DC subsets within total CNS DCs. (C) Representative immunofluorescence images of CD11c+ MHCII+ CD11b cDC1s and CD11b+ cDC2s in the dura mater, leptomeninges, and choroid plexus of the steady-state brain (n ≥ 2 mice, ≥4 sections per mouse). Scale bar, 50 μm. A zoomed-in view is shown. (D) Brain compartments [leptomeninges, dura mater, choroid plexus, and parenchyma (hippocampus)] were dissected from n = 10 brains. Individual compartments were pooled and single cells were analyzed by flow cytometry followed by computational high-dimensional data analysis. After gating on live singlets, CD45+ cells were exported as an FCS file using the FlowJo software. Shown are pooled data from n = 10 mice, visualized using t-SNE and clustered using the FlowSOM algorithm in R (142,655 total cells; max iterations = 10000, perplexity = 50, θ = 0.5). Plots show FlowSOM clusters of leukocytes overlaid onto a t-SNE map of the combined dataset (left) or shown separately for each compartment as indicated. (E) Data show frequencies of cDC1s, cDC2s, and pDCs of total DCs (left graph) or CD45+ cells (right graph) as determined by FlowSOM clustering. (F) DCs (CD11chi, MHCII+, B220, MerTK), BAMs (MerTK+, CD11b+, CD45hi, MHCII+), microglia (MerTK+, CD45lo, CD11b+), and B cells (CD11b, B220+, CD11c, MHCII+) were isolated from pooled whole steady-state brains of 20 to 25 C57BL/6 mice and incubated with in vivo activated encephalitogenic 2D2 cells in the presence [2000:55,000 (APC:T cell); upper graph] or absence [11,000:55,000 (APC:T cell); lower graph] of MOG35–55 peptide for 72 hours. IFN-γ production was measured by ELISPOT assay. Data represent the mean ± SD cytokine activity of IFN-γ from technical duplicates and are representative of one of two independent experiments.

Despite their critical role in T cell reactivation, cDCs are sparsely distributed in the steady-state CNS. Using immunohistochemistry, we localized CD11c+ MHCII+ cells [which almost exclusively consist of DCs in the steady-state brain (fig. S4B)] primarily in the leptomeninges and dura mater with fewer cells in the choroid plexus (Fig. 4C), whereas DCs were largely absent from the perivascular spaces, confirming previous observations (57). Early T cell encounters with brain-derived Ags specifically occur in the parenchyma surrounding leptomeningeal spaces (consisting of the pia and arachnoid mater), highlighting the strategic localization of DCs to reactivate CNS-infiltrating T cells (15, 20). We confirmed that DCs in the leptomeninges and dura mater consisted mainly of CD11b+ cDC2s, whereas CD11b cDC1s were the dominant DC subset in the choroid plexus (Fig. 4C). To quantify the relative distribution of cDCs across different CNS compartments, we physically separated the dura mater, leptomeninges, choroid plexus, and parenchyma/perivascular spaces and analyzed the subset composition of each compartment using flow cytometry. This approach is only an approximation because of potential cross-contamination of different brain compartments (mainly of the leptomeninges with parenchymal tissue) and because it is not possible to isolate the complete leptomeninges of adult mice. However, using algorithm-guided clustering (FlowSOM), we found that cDC2s make up to 80% of brain DCs in the dura mater, the pia mater, and the parenchyma/perivascular spaces (here, we only found very low numbers of CD11c MHCII+ DCs). In contrast, the choroid plexus revealed a dominant presence of the cDC1 subset (Fig. 4D and fig. S4C). This was also confirmed using a manual gating strategy to quantify cDC1 and cDC2 cells (fig. S4D). Together with the rarity of cDC1 cells, that Batf3−/− mice, which lack cDC1s in the CNS (fig. S4, E and F), were fully susceptible to transfer EAE, provides an additional argument against a role of cDC1s in the reactivation process of encephalitogenic CD4+ T cells.

cDCs process and present myelin Ag to infiltrating T cells

Last, to independently compare the ability of CNS APCs to reactivate infiltrating MOG35–55-specific T cells in vitro, we sorted cDCs, BAMs, B cells, and microglia from the steady-state brain of adult C57BL/6 mice (fig. S4G) and incubated them with in vivo–primed 2D2 CD4+ T cells in the presence or absence of MOG35–55 peptide. In the presence of exogenous peptide, cDCs induced the strongest activation of T cells in terms of interferon-γ (IFN-γ) expression as measured by enzyme-linked immunospot (ELISPOT) assay (Fig. 4F). The second most potent APCs were BAMs, whereas B cells and microglia failed to induce IFN-γ production in primed 2D2 cells. We observed that DCs could reactivate myelin-specific 2D2 T cells even in the absence of exogenous peptide (Fig. 4F). No other APC was able to reactivate primed CD4+ T cells without the addition of MOG35–55 peptide in vitro. These findings demonstrate that CNS cDCs actively process and present endogenous MOG protein in the steady-state CNS and confirm their superior Ag presentation capacity compared with other APCs (6).


Active communication between the CNS and the immune system has a profound impact on tissue maintenance and pathology. Yet, even already primed TH cells are relatively innocuous when they do not reencounter their cognate Ag in the respective target tissues. The nature and origin of the APCs required to reactivate CNS-specific, encephalitogenic TH cells has been heavily disputed because the strong overlap in CNS-associated myeloid cell phenotypes has hindered deciphering their individual contributions in this critical process (7, 15, 3335). We here present a precise categorization of MHCII-expressing cells in the steady-state CNS and a systematic study of their individual ability to elicit inflammatory immune responses in the CNS.

EAE has served for decades as a model for autoimmunity, cellular immune activation, T cell–APC interactions, and cytokine networks in inflammation (36). The AT EAE model resembles several aspects of MS, as it mimics immune infiltration into the healthy CNS in the absence of systemic inflammation. Because the transient contact between adoptively transferred encephalitogenic T cells with stromal cells in the periphery is MHCII independent (7, 37), our investigation of the MHCII-dependent T cell reactivation could be restricted to the CNS, rather than other peripheral organs. Crossing the endothelial cell barrier from the vessel lumen to the meninges by encephalitogenic T cells occurs in an Ag-independent manner (20, 37), whereas the presentation of myelin epitopes by CNS APCs in the leptomeningeal space is required for their entry into the CNS parenchyma and for the initiation of disease (20).

Astrocytes, pericytes, and endothelial cells have been suggested as possible APCs in the CNS (14). In the non-inflamed brain, however, most cell types (particularly non-immune cells) do not express MHCII or costimulatory molecules and are therefore unlikely to be responsible for the initial reactivation of encephalitogenic T cells. In addition, microglia, the only leukocytes of the steady-state brain parenchyma, have been implicated to promote the pathogenesis of EAE (38). We identified a minor MHCII+ microglial cell population (up to 3% of total microglia) in the steady-state CNS. However, although microglia have the capacity to process and present exogenous Ag onto MHCII molecules in principle, their anatomical location is distant from the CNS interfaces, where T cells make initial contact with the brain. Using Sall1CreERT2/+ and Cx3cr1CreERT2/+ Iabfl/fl mice, we confirmed that microglia do not have an essential role in initial T cell reactivation in AT EAE (7, 26).

Considering the relative abundance of bona fide MHCII-expressing cells in the steady-state CNS interfaces (pia mater, choroid plexus, and perivascular spaces) implicated in T cell reactivation (15, 20, 21), we considered BAMs, cDCs, and B cells as the main candidate APCs to reactivate T cells in EAE. In MS, B cells recently have been shown to induce autoproliferation in brain-homing T cells in an human leukocyte antigen (HLA)-DR–dependent manner (39); however, for the reactivation of CNS-infiltrating encephalitogenic T cells in the MOG35–55-dependent EAE model, they appear to be entirely dispensable (40). We here confirmed that even with external MOG35–55 loading, CNS-associated B cells are relatively poor APCs to myelin-specific T cells. The most hotly debated CNS APCs are BAMs and DCs (7, 15, 4143). Similar to microglia, most CNS BAMs are embryonically derived cells (4, 8), and we found that most of those cells do not express MHCII in the steady-state brain (5). However, others have observed that macrophage depletion in EAE can reduce clinical progression by blocking parenchymal T cell invasion (41). BAMs are strategically located at CNS-immune interfaces (5, 8, 15) as ideal candidates to present Ag to incoming T cells. However, “meningeal macrophages,” as described by Schläger et al. (15), were so named on the basis of their morphology and ability to take up dextran molecules. We found morphology alone not to allow the unambiguous discrimination of BAMs from other members of the mononuclear phagocyte system, most notably cDCs (5). Compared with CNS BAMs, DCs are a very rare population and, as such, have often evaded detection in the steady-state CNS. It is likely that the steady state function of these rare CNS DCs is tolerance maintenance. Global CD11c+ cell depletion was shown to exacerbate the clinical outcome of actively induced EAE (34, 44).

In contrast, tolerance is bypassed in AT EAE by the infusion of primed, encephalitogenic T cells, which, in turn, must encounter their cognate neuro-Ag in the tissue. The threshold for reactivation of already primed T cells is much lower, and the cognate TCR/MHCII peptide encounter is dominant (45). AT of encephalitogenic T cells into DC-depleted recipients failed to trigger disease (33, 46). Moreover, DCs have been proposed previously to be sufficient to reactivate CNS-infiltrating T cells when MHCII expression was confined to CD11c+ cells (7). However, in vivo cell depletion often results in collateral inflammation and alteration of tissue homeostasis because other cells often invade the empty cellular niche (47). In addition, CD11c can no longer be considered a DC-specific marker (1113, 38). Hence, a more systematic interrogation of APC capacity across the CNS-leukocyte landscape was necessary.

New insights into the differences between brain-associated myeloid cells in terms of ontogeny and longevity made it possible for us to exploit the Cx3cr1CreERT2/+ Iabfl/fl system to differentially target MHCII+ expression in macrophages and DCs by using different tamoxifen treatment regimens. Using this strategy, we revealed that cDCs, but not embryonically derived BAMs and microglia, are required for initial T cell reactivation and parenchymal infiltration. Furthermore, we showed that cDCs isolated from the CNS of naïve C57BL/6 mice are superior in presenting MOG35–55 Ag to TCR transgenic primed 2D2 cells when compared with microglia, B cells, or MHCII+ BAMs. We also found clear evidence for myelin sampling, processing, and presentation by CNS cDCs, when freshly isolated CNS cDCs were capable of triggering specific T cell reactivation even in the absence of pulsed peptides. In line with in vivo imaging data in rodent EAE models, there is mounting evidence to support that meningeal inflammation precedes the clinical manifestation of MS (48, 49). Whereas the choroid plexus is mainly home to CD11b cDC1s, the cDC2 subset is enriched in the CNS meninges (57, 33). Most of the initial CNS-invading T cells accumulate in the leptomeninges rather than the choroid plexus (15, 20). This is in line with the view that cDC2 cells, which are found in the leptomeninges and not in the choroid plexus, are the main facilitators of early T cell entry into the CNS.

In summary, our data demonstrate that parenchymal microglia do not serve as early APCs to myelin Ag–restricted T cells. Further, despite their relative abundance and ideal location at the CNS-immune interface, BAMs are unable to process and present myelin Ag to arriving T cells. By contrast, and despite their scarcity, CNS-associated cDCs, in particular the leptomeningeal-associated cDC2 subset, are exquisite and powerful APCs and the only bridging cellular element that can permit effective T cell–CNS interactions.


Study design

This study aimed at identifying and characterizing CNS-resident APCs in the steady state and their ability to sample myelin Ag for presentation to T cells. For this, we used a combination of multiparametric single-cell mass cytometry and algorithm-guided computational analysis. To then interrogate the ability of different CNS APC populations to engage with the first-arriving encephalitogenic T cells, we used Cre-loxP–mediated MHCII ablation during AT EAE in vivo.


2D2 TCR transgenic mice [C57BL/6-Tg(Tcra2D2,Tcrb2D2)1Kuch/J; #006912] (24), Lyz2Cre mice [B6.129P2-Lyz2tm1(Cre)Ifo/J; #004781] (27), ItgaxCre mice [B6.Cg-Tg(Itgax-Cre)1-1Reiz/J; #008068] (28), ItgaxCreEGFP mice [C57BL/6 J-Tg(Itgax-Cre,-EGFP)4097Ach/J; #007567] (29), Zbtb46Cre mice [B6.Cg-Zbtb46tm3.1(Cre)Mnz/J; #028538] (30), Iabfl mice (B6.129X1-H2-Ab1tm1Koni/J; #013181) (23), and Batf3−/− mice [B6.129S(C)-Batf3tm1Kmm/J; #013755] (50) were purchased from the Jackson Laboratory. Cx3cr1CreERT2 mice [B6.129P2(C)-Cx3cr1tm2.1 (Cre/ERT2)Jung/J; #020940] (31) were provided by S. Jung (Weizmann Institute of Science, Israel), and Sall1CreERT2 and Sall1GFP mice (51) were provided by R. Nishinakamura (Kumamoto University, Japan). Transgenic mice were bred in house, and C57BL/6 mice were purchased from Janvier. Age- and sex-matched (male and female) 6- to 12-week-old mice were used for all experiments. All experiments performed in this study were approved by the Swiss Veterinary Office and performed according to federal and institutional guidelines.

Tamoxifen treatment

Tamoxifen (Sigma) was dissolved in ethanol and corn oil (1:9) to 25 mg/ml and administered in 200-μl doses via oral gavage (5 mg per dose).

Adoptive transfer EAE

Donor mice (2D2) were immunized with 200 μg of MOG35–55 emulsified in CFA and treated with 200 ng of pertussis toxin (intraperitoneally) on the day of immunization and again 2 days later. After 9 days, splenocytes and lymph node (inguinal, axillary, and brachial) cells were cultured in complete RPMI (10% fetal calf serum, penicillin/streptomycin, and β-mercaptoethanol) containing MOG35–55 (20 μg/ml; GenScript), recombinant IL-23:Fc fusion protein (10 ng/ml; ACROBiosystems), and anti–IFN-γ (5 μg/ml; R4-6A2; BioExpress) at a density of 6 × 106 to 8 × 106 cells/ml for 72 hours. To induce EAE, we injected (intraperitoneally) up to 20 × 106 cells into recipient mice.

Tissue preparation and single-cell suspension

In brief, tissues (whole brain and spinal cord) were cut into small pieces and incubated with collagenase type IV (0.4 mg/ml) and deoxyribonuclease I (DNase I) (0.2 mg/ml) (Sigma-Aldrich) for 30 to 40 min and passed through a 19-gauge needle to obtain a homogeneous single-cell suspension. CNS cell suspensions were further enriched by Percoll gradient (30%) centrifugation (1590g, 4°C, 30 min, no brake).

Individual brain compartments—including leptomeninges, dura mater, choroid plexus, and parenchyma (hippocampus without leptomeninges or choroid plexus)—were microdissected. The dura mater was removed from the skull after cortex removal. The leptomeninges was partially removed from the cortex by starting at the olfactory bulb and gently peeling off the soft meningeal layer. The choroid plexus was dissected from the fourth ventricle by separating the cerebellum from the cerebral cortex. Next, the two cerebral hemispheres were gently separated, and the choroid plexus from the lateral ventricles and the third ventricle was harvested. To exclude contamination of residual leptomeninges on the surface of the cortex, we dissected the hippocampus from both hemispheres after the choroid plexus was removed. The hippocampus was used as a control for the parenchyma and perivascular spaces because it only harbors blood vessels without contamination by leptomeninges or choroid plexus. All brain compartments were separately subjected to collagenase type IV/DNase I digestion (see above). CNS leukocytes from the parenchymal hippocampus samples were further enriched by Percoll gradient centrifugation (see above).

Flow cytometry

Flow cytometry was performed on an LSRII Fortessa (Becton Dickinson) and analyzed with FlowJo software (Tree Star). Fluorochrome-conjugated monoclonal antibodies specific for mouse MHCII I-A/I-E (clone M5/114.15.2), CD11b (clone M1/70), CD11c (clone N418), CD45 (clone 30-F11), CD24 (clone M1/69), Ly6G (clone 1A8), Ly6C (clone HK1.4), F4/80 (clone CI:A3-1), CX3CR1 (clone SA011F11), MerTK (DS5MMER), CD4 (clone GK1.5), Flt3 (clone A2F10), CD31 (clone 390), Lyve1 (clone ALY7), XCR1 (clone ZET), B220 (clone RA3-6B2), GR-1 (clone RB6-8C5), and CD206 (C068C2) were purchased from either BD Biosciences, eBioscience, or BioLegend. Dead cells were excluded using a LIVE/DEAD stain kit (Invitrogen), and doublets were excluded by FCS-H versus FSC-A in FlowJo.

Ex vivo Ag presentation assay

DCs (CD11chigh, MerTK, MHCIIhi), MHCII+ BAMs (MerTK+, CD11b+, MHCII+), microglia (MerTK+, CD45low, CD11b+), and B cells (B220+, MHCII+) were sorted by FACS and incubated with activated (day 9 active EAE) magnetically sorted (negative selection, CD4+ T Cell Isolation Kit from Miltenyi Biotec) 2D2 CD4+ T cells in the presence or absence of MOG35–55 peptide (20 μg/ml) at 37°C for 72 hours. Cells were analyzed for IFN-γ expression by ELISPOT (Abcam).

Mass cytometry

Mass cytometry experiments were performed exactly as recently described (5). Mass cytometry antibodies were either labeled in-house using antibody-labeling kits and protocols or purchased from Fluidigm. Antibodies were individually titrated and optimized for the final panel before use to ensure that each parameter was informative. For MHCII+ population identification, data were taken from Mrdjen et al. (5). Here, five palladium metal isotopes were used for live cell barcoding of samples with CD45 while keeping CD45-147Sm as a common channel to identify cells expressing varying degrees of CD45, such as microglia (5). Briefly, cells from individual naïve adult C57BL/6 brains were incubated with respective CD45-Pd + CD45-147Sm antibodies in phosphate-buffered saline (PBS) for 30 min at 37°C, after which they were washed twice with FACS buffer (PBS supplemented with 0.5% bovine serum albumin) and then combined into composite samples. For DC subset analysis, we used a DC-focused antibody panel with additional DC markers. Here, brain samples were prepared as described and pooled without barcoding. After pooling, cells were enriched by FACS to remove the majority of non-DC populations. Cells were stained with the cocktail of primary CyTOF antibodies for 30 min at 37°C, washed with FACS buffer, and then incubated with secondary antibodies for 20 min at 4°C. After washing, samples were incubated with intercalating solution (iridium from Sigma) in Maxpar Fix/Perm buffer (Fluidigm) overnight at 4°C. Before acquisition, the samples were washed twice with FACS buffer and once with Milli-Q water. Barcoded composite samples were acquired on a Helios mass cytometer (Fluidigm). Quality control and tuning processes on the Helios were performed by routine before acquisition.

Preprocessing of mass and flow cytometry data

Cytometry data were processed as previously described (5, 18). Mass cytometry data were normalized to EQ Four Element Beads (Fluidigm), and live cells were exported by gating on event_length, DNA (191Ir and 193Ir positive), and live cells (195Pt negative) using FlowJo software (Tree Star). For the barcoded samples, debarcoding was achieved by Boolean gating in FlowJo. After quality control for each channel, the debarcoded gates or complete pooled sample FCS files were imported into the R environment and transformed using an inverse hyperbolic sine (arcsinh) function with a cofactor of 5.

For flow cytometry data, after compensation correction in FlowJo, live, single, quality-controlled, and compensated cells were exported by manual gating. The appropriate transformation cofactors were determined by uploading the files into Cytobank ( and using the Scales feature; thereafter, transformation was carried out in MATLAB, and transformed files were imported into the R environment for further preprocessing and analysis. To equalize the contribution of each marker in subsequent automated data analysis steps of both mass and flow cytometry data, we performed percentile normalization, normalizing all data to the 99.9th percentile of the merged sample in each experiment, depending on the number of outliers present. This process preserves biologically relevant differences in expression values (staining indices) but normalizes intermarker maximum and minimum expression values from 0 to 1.

Automated population identification in high-dimensional data analysis

Preprocessing of the raw data was followed by dimensionality reduction and visualization by t-SNE (17). Initial clustering and meta-clustering with FlowSOM (19) was followed by expert-guided manual merging of some of the metaclusters using the information from the t-SNE with overlaid marker expression values and a heatmap of median expression values (18).

Statistical analysis

For clinical scores over time, influence on MHCII expression in different cell populations, or analysis of CNS infiltration of different cell types between different groups, differences were evaluated by two-way analysis of variance (ANOVA) with Bonferroni’s post hoc test, Tukey’s or Sidak’s multiple comparisons. Differences for one parameter were evaluated by one-way ANOVA with Bonferroni’s post hoc test when comparing more than two groups and by the two-tailed Student’s t test when comparing two groups (unpaired t test with Welch’s correction when the two samples have unequal variances and unequal sample sizes). χ2 test was applied to compare incidence of EAE. P values of less than 0.05 were considered statistically significant, with *P ≤ 0.05, **P ≤ 0.01, and ***P ≤ 0.001.


After transcardiac perfusion of mice with PBS and 4% (w/v) paraformaldehyde (PFA) in 0.1 M phosphate buffer (pH 7.4), CNS tissue was postfixed overnight at 4°C, followed by cryoprotection with 30% (w/v) sucrose in PBS. Sections were cut at 14 μm and mounted on slides, or free-floating sections were cut at a thickness of 30 μm using a Hyrax C60 cryostat (Zeiss). CNS tissue sections were permeabilized by incubation in blocking solution (PBS supplemented with 0.1% Triton X-100 and 4% normal goat serum) for 30 min to 1 hour at room temperature. Subsequently, sections were incubated with the following primary antibodies (diluted in blocking solution) for 24 to 72 hours at 4°C: anti–GLUT-1 (1:500; Millipore), anti-Iba1 (1:500; Wako), anti-MHCII (1:200; BioLegend), anti-CD206 (1:100; BioLegend), anti-CD11c (1:40; BioLegend), and anti-CD11b (1:50; BioLegend). After washing, samples were incubated either at 4°C overnight or at room temperature for 1 to 2 hours with the respective Alexa Fluor–conjugated secondary antibodies (1:500; Life Technologies). Counterstaining was performed using SlowFade Gold antifade reagent with 4′,6-diamidino-2-phenylindole (DAPI) (Invitrogen). Fluorescence photomicrographs were captured with a confocal laser scanning microscope (SP5; Leica, Heerbrugg, Switzerland) equipped with argon and helium lasers using a 40× objective (oil immersion; numerical aperture, 1.25). Images were processed and merged using Imaris imaging software (Bitplane, Zurich, Switzerland).

Histopathological analysis of EAE tissue

LFB (Luxol fast blue)–PAS (periodic acid–Schiff) stainings were done according to standard protocols on 4% PFA-fixed, decalcified paraffin-embedded 6-μm-thick tissue sections. We analyzed EAE histopathology on cross sections (four to six per mouse) of lumbar spinal cords and recorded digital images of tissue sections with a light microscope (Olympus BX41). Total demyelinated area of LFB- and PAS-stained sections was measured by Fiji/ImageJ version 1.46j software (National Institutes of Health), and the area of demyelination was calculated as the percentage of the whole area of white matter within a given section.


Fig. S1. Identification of MHCII+ cells in the steady-state CNS.

Fig. S2. Microglial MHCII is dispensable for AT EAE.

Fig. S3. cDCs, but not BAMs or microglia, are required for the reactivation of 2D2 CD4+ T cells in the CNS.

Fig. S4. Different DC subsets reside at the steady-state brain interfaces and present myelin Ag to CD4+ 2D2 T cells.

Table S1. Raw data.


Acknowledgments: We thank the Cytometry Facility (University of Zurich) and the Center for Microscopy and Image Analysis (University of Zurich) for technical assistance, Insight Editing London for critical review and editing of the manuscript, and N. Puertas for excellent technical assistance. Funding: This work was supported by grants from the Swiss National Science Foundation (PP00P3_144781 to M.G. and 310030_146130, 316030_150768, and 310030_170320 to B.B.), the European Union FP7 ITN_NeuroKine (to B.B.), the European Union FP7 Project ATECT (to B.B.), and the University Research Priority Project Translational Cancer Research (to B.B.). Author contributions: S.M. and B.B. conceived the study. B.B. supervised the study, and S.M. performed all experiments and statistical analysis with support from D.M. for mass cytometry experiments, D.M. and S.G.U. for FACS, and B.S. and S.G.U. for immunohistochemistry experiments. S.M. and B.B. wrote the manuscript with critical review from D.M., S.G.U., B.S., and M.G. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The CyTOF data have already been made public under (5). The CyTOF data shown in Fig. 4 are available at

Stay Connected to Science Immunology

Navigate This Article