Classic flow cytometric gating fails to discriminate resident from infiltrating myeloid cells during severe inflammation
During viral encephalitis, the overlapping expression of cell surface markers makes the identification of resident and infiltrating myeloid cells in the CNS difficult by standard gating approaches. We therefore set out to identify the minimum parameters required to delineate these populations accurately. Cells were dissociated from murine brains at various timepoints in the progression of lethal WNE. Figure 1a-c shows a clear CD45lo, CD11bhi microglial cell population which expresses low levels of Ly6C in the brain parenchyma of dpi 0 mice. However, from dpi 3 onwards this discrete microglial cell population became progressively obscured by the overlapping antigen expression of increasing numbers of CD45hi, CD11bhi, Ly6Chi infiltrating BM-derived monocytes. From dpi 5–7 these monocytes showed further CD45 and CD11b upregulation with some downregulation of Ly6C (Fig. 1a-d). Compounding this, by dpi 5, microglia had also upregulated CD45 and Ly6C (Fig. 1a, b), making it impossible to separate these populations accurately by standard gating during WNE.
To resolve this, we used the ‘microglia-specific’ markers, Transmembrane Protein 119 (TMEM119) (36) and purinergic receptor P2Y12 (P2RY12) (10). However, while collagenase/DNase treatment substantially increases leukocyte yields and live:dead cell ratios during preparation of single cell suspensions (Additional file 1), TMEM119 did not label cells prepared this way (Fig. 1e, f). This has not previously been reported, to our knowledge. Ironically, without enzyme treatment, TMEM119 was downregulated on microglia at dpi 7, precluding the use of this microglia-specific marker to distinguish between resident and infiltrating myeloid cells (Fig. 1g). On the other hand, P2RY12 was unaffected by enzyme treatment, but was expressed both by a discrete CD45int/lo population at day 0 and on different populations of CD45int/hi cells at dpi 7 (Fig. 1h). Thus, it was also unclear whether this purinergic receptor, putatively expressed only by microglia, was also expressed by an infiltrating CD45hi macrophage population. It should be noted that collagenase/DNase was used in the remainder of this study for optimal brain cell preparation, except where TMEM119 was measured.
High parameter cytometry and dimensionality-reduction can delineate resident and infiltrating myeloid cells
As shown above, the discrimination of microglia from infiltrating myeloid cells in severe neuroinflammatory conditions is unreliable, even with the use of ‘microglia-specific’ markers. Other groups have distinguished microglia from macrophages based on the higher macrophage expression of CD44 (37, 38), VLA-4 (39), CD11a (40), CCR2 and Ly6C. In WNE, the expression profile of these markers on some infiltrating myeloid cells, viz., Ly6ChiCD45int and Ly6CintCD45int was similar to microglia in the infected brain (Additional file 2). Thus, we were also unable to distinguish these populations using these markers. To address this issue, we labelled cells for myeloid cell markers (CD45, CD11b, F4/80), ‘activation/functional’ myeloid markers (CD11c, MerTK, CD64, CD68 CD86, MHC-II), ‘infiltrating/inflammatory macrophage’ markers (CCR2, Ly6C) and markers ‘specific’ for and/or highly expressed on microglia (P2RY12, TMEM119, CX3CR1) and analysed them by fluorescence flow cytometry. We visualized the acquired high parameter data on a 2D plot after subjecting it to T-distributed Stochastic Neighbor Embedding (tSNE), which clusters cell populations based on similarity of marker expression (41). We then generated a series of gating strategies to identify discrete populations clustered on the tSNE plot.
In the homeostatic brain four different fluorescence gating approaches effectively identified the same microglial population (Additional file 3). However, in severe neuroinflammation (Fig. 2) these approaches produced different results. Gating strategy 1 identified ‘resting’ microglia as CD45loCD11b+ and ‘activated’ microglia as CD45intCD11b+ (Figs. 2f). This strategy has typically been used to identify microglia in both the homeostatic and diseased brain (1, 42, 43). The CD45loCD11b+ ‘resting’ microglia gate comprises a single cluster when overlaid onto a tSNE plot from WNV dpi 7 brains (Fig. 2g, blue). However, the CD45intCD11b+ gate captures infiltrating monocytes/macrophages in neuroinflammatory and neurodegenerative models (1, 44). Accordingly, the ‘activated’ microglia gate comprises 2 distinct clusters on the tSNE plot (Fig. 2g, red).
We therefore generated and compared further gating strategies to determine the feasibility of more accurately distinguishing resident from infiltrating myeloid populations (Fig. 2). Microglia were identified as Ly6G− CX3CR1+ CD45lo/int CD11b+ Ly6C−/lo (Gating strategy 2, Fig. 2j, k), Ly6G−, CX3CR1+ CD45lo/int CD11b+ P2RY12+ Ly6C−/lo (Gating strategy 3, Fig. 2l, m) and Ly6G−, CD45int/lo, PR2Y12+, CX3CR1hi − lo, CD11b+ (Gating strategy 4, Fig. 2o, p). Strategies 2 and 3 resulted in the inclusion of various populations outside of the putative microglial cluster in the tSNE plot, while strategy 4 identified a single microglial cluster with very little contamination from other cells. Strategy 4 used the fewest markers and gates, the simplest gating, and was thus least susceptible to error. We employed gating strategy 4 to further investigate the identities and change kinetics of these populations during WNE. Nevertheless, since strategies 2 and 4 gave putative microglial numbers that were not statistically different (Figs. 2q, r), previous data using standard markers prior to the availability of P2RY12 could be re-analysed more accurately using strategy 2.
Modulating CNS infiltration verifies non-microglial populations in Gating strategy 4.
In order to confirm that Gating strategy 4 was accurately distinguishing resident from infiltrating myeloid populations during WNE, we employed 3 approaches. We 1) blocked the entry of myeloid populations infiltrating into the brain from peripheral blood, 2) adoptively transferred BM-derived monocytes and tracked their phenotypic changes after infiltration and 3) injected intravenous dye to label peripheral leukocytes that infiltrate into the brain.
Infiltration of cells into the brain parenchyma was blocked using an antibody cocktail made up of of anti-VLA-4 (3), CCL2 (1) and Ly6C (45) (Fig. 3a) at dpi 5 and 6. This resulted in a 90–95% reduction in the number of infiltrating Ly6Chi inflammatory and a population of ‘microglia-like’ macrophages (i.e., non-microglial myeloid cells present in the infected brain with a CD45+, CX3CR1+ and CD11b+ profile similar to microglia, but absent in the homeostatic brain). This was associated with a commensurate depletion in the corresponding areas of the tSNE plot. Importantly, the antibody cocktail did not reduce numbers of putative microglia in animals treated with blocking antibodies, compared to untreated animals (Fig. 3b-d). This suggests that Gating strategy 4 identifies microglia, even during highly inflammatory conditions.
Unexpectedly, the use of the isotype control antibody cocktail increased the number of microglia and infiltrating macrophages compared to untreated mice (Fig. 3d), raising the question of whether some of the infiltrating cells were falling into the putative microglia gate. To investigate this, we adoptively transferred CFSE+ BM-derived CD115+, CD45+, CD11b+, Ly6Chi, Ly6G−, B220− monocytes from dpi 5 WNV-infected mice, into recipient age-, sex- and time-matched WNV-infected animals and harvested the recipient brains on dpi 7 (Fig. 3e). Analysis shows that all of the transferred cells fell outside the identified putative resident microglia tSNE cluster (Fig. 3f-h) and appeared in both the infiltrating Ly6Chi inflammatory and ‘microglia-like’ macrophage gates (Fig. 3i).
Considering only a small number of transferred cells could be tracked via adoptive transfer, we also injected mice with PKH67 three hours prior to collecting brain tissue, to definitively distinguish infiltrating from resident myeloid populations (Fig. 3j, k). At dpi 5, 6 and 7, the majority of PKH67+ cells infiltrating WNV-infected brains were infiltrating monocytes or lymphocytes, while cells in the putative microglial cluster showed no staining. Taken together, the data strongly suggests that this cluster only represents the resident microglia.
Microglia adopt ‘disease-specific’ immunophenotypes in WNE
Using Gating strategy 4, we then proceeded to investigate the immunophenotypic heterogeneity and response of microglia during WNE. We pre-gated live, CD11b+, CD45+ myeloid cell populations and ran tSNE analysis on the concatenated dpi 7 WNV-infected and mock-infected populations (Fig. 4a-c).
Clustering the pre-gated cells on CD45, CD11b, CX3CR1, CD64, F4/80, CD86, CD68, P2RY12, Ly6C, CD44, MHC-II, CD11c, CCR2, MerTK and TMEM119, revealed 4 distinct microglial phenotypes in both the homeostatic and infected brain (Fig. 4b, c). These were, PR2Y12hiCD86−, PR2Y12hiCD86+, PR2Y12loCD86− and PR2Y12loCD86+. Importantly, Gating strategy 4 could identify these subsets in enzyme- and non-enzyme-digested brains, as this approach does not rely on TMEM119 or CD44 detection, which is reduced on digestion (Additional file 4).
Strikingly, the microglial cluster in the homeostatic brain was in a different position on the tSNE plot from that in the infected brain. This indicates a substantial phenotypic change in microglia during infection (Fig. 4a-c). Consistent with this, at dpi 7 all microglia phenotypes had downregulated TMEM119, CX3CR1, F4/80 and CD68 and upregulated CD45 and CD64. (Fig. 4d-f and Additional file 5). However, in contrast to the significant downregulation of TMEM119, P2RY12 expression was relatively stable from dpi 0 to 7 (Fig. 4d-h). Notwithstanding these phenotypic changes, the same four distinct subsets could be identified throughout the disease course.
At dpi 7, the P2RY12hi microglia subsets in both mock-infected and infected brain had higher expression of all measured markers, compared to the P2RY12lo subsets (Fig. 4f-h). P2RY12hi cells from infected brains upregulated CD45, CD11b, CD64, MerTK, CD11c, CD44, CCR2 and Ly6C, relative to microglia at dpi 0. Of the P2RY12hi population, the CD86+ microglia showed the highest expression of CX3CR1, F4/80, TMEM119, CD68, MerTK, MHC-II, CD44, CCR2 and Ly6C, while CD86− microglia showed the highest expression of CD11b, CD64 and CD11c. Notably, microglia that expressed the highest levels of ‘activation’ markers in the infected brain paradoxically also had the highest expression of nominally homeostatic markers, CX3CR1, TMEM119 and P2RY12.
In contrast to microglia, infiltrating myeloid cells in these tSNE plots fell into 2 principal populations, Ly6Chi inflammatory macrophages (CD45hi, CD11bhi, CX3CR1lo, F4/80+, CD64hi, CD68hi, CD44hi, Ly6Chi, 72.5%) and microglia-like macrophages (CD45int, CD11bint, CX3CR1hi/int, F4/80+, CD64int, CD68hi, CD44int, Ly6Cint, 23.5%) (Fig. 4e). Both populations had varied expression of MerTK, CD86, CD11c, MHC-II and CCR2, as well as TMEM119 and P2RY12. Since peripheral myeloid populations in the blood and bone marrow expressed neither TMEM119 nor P2RY12 (Additional file 6 and 7), this indicates that myeloid cells upregulated these ‘microglia-specific’ markers, de novo after CNS infiltration.
Microglial phenotypes depleted by PLX5622 are not brain region-specific
Abrogating myeloid and lymphoid cell infiltration using systemic antibody blockade, did not reduce microglial numbers (Fig. 3d, 5a, b). In contrast, the 4 phenotypes defined by CD86 and P2RY12 expression were substantially and proportionally decreased in animals treated with chow containing CSF1R inhibitor, PLX5622. (Fig. 5c, d). Taken together, these data strongly suggest that these cells are resident microglia.
There have been limited reports of a CD86-expressing microglial population in the homoeostatic or infected brain. Therefore, we first established that the expression of CD86 was not a result of the reported non-specific binding of cyanine dyes (APC-Cy7 – Fig. 4) to macrophages (Additional file 8). To further confirm this, we examined brain sections for a CD86+ microglia using metal-labelled antibodies in imaging mass cytometry (IMC). At dpi 5 with minimal myeloid cell infiltration, microglia were readily identifiable by their ramified morphology and Iba1+ and Ly6C− expression profile. These cells were CD86+ in various regions of the brain (Fig. 5e, f).
In dissecting the brain into 5 separate regions, viz., olfactory bulb, frontal cortex, posterior cortex, pons/medulla and cerebellum, the 4 identified microglial phenotypes were found in all areas by flow cytometry (Fig. 5g). This indicates that these phenotypes are not region-specific and supports the IMC data. Nevertheless, there was significant variation in the proportion of each microglial subset between these anatomical areas, both under homeostatic conditions and in response to WNV infection (Fig. 5h, I, 6b). Under homeostatic conditions, the largest group was the CD86− microglia, with the P2RY12hi subset comprising 68–80% and the P2RY12lo subset, 13–28%. The CD86+ subsets together comprised less than 5% of microglia. In response to WNV, there was an increase in the proportions of both P2RY12lo subsets in all anatomical areas by dpi 7, principally at the expense of the P2RY12hiCD86−(Fig. 5h, i).
Microglia in the olfactory bulbs had the highest expression of CX3CR1, F4/80, CD68 and MHC-II in both the homeostatic and infected brain. However, in WNV-infected brains they had the lowest expression of CD11b, TMEM119 and P2RY12 of all the anatomical sites (Fig. 6a-b). Considering WNV enters the olfactory bulb and remains there over the course of infection, downregulation of these markers in this model could be a result of prolonged neuronal infection and exposure to neuroinflammation. Downregulation of TMEM119 and P2RY12 has been reported in a number of chronic neuroinflammatory models (46–48).
Considering the demonstrable progression of infection from rostral to caudal over time, it was of interest to determine changes in marker expression on microglia in the cerebellum. The cerebellum shows limited neuronal infection at dpi 7, despite a broad interferon-stimulated gene response (49). Cerebellar microglia had a higher CD11b and MerTK expression and lower CX3CR1, F4/80, CD68 and MHC-II expression than the olfactory bulb. Nevertheless, proportions of both P2RY12lo subsets were similarly increased in response to WNE in both regions (Fig. 5h, i).
More striking was the differential profile of Ly6Chi macrophages in these brain regions at WNV dpi 7 (Fig. 6c, d). From caudal to rostral, infiltrating Ly6Chi macrophages had progressively downregulated Ly6C and upregulated CX3CR1, TMEM119, P2RY12, CD64, CD68 and MHC-II. Resident and infiltrating cells in all regions remained distinct, clustering in separate groups on the tSNE plot. However, they were clustered more closely together rostrally, indicating a phenotype closer to microglia in the olfactory bulb (Fig. 5g, 6c).
Temporal changes in microglial phenotypes during WNE
The microglial phenotypes we identified in the naïve and WNV dpi 7 brain were also present at dpi 4–6 of WNE. With increasing infection in the brain, all microglia showed significant temporal phenotypic changes (Fig. 7a-e). Measured markers were progressively a) upregulated, b) downregulated or c) upregulated and then downregulated, indicating changes in potential activity and functions at different disease stages (Fig. 7a, c-e).
From dpi 4, the total microglia population upregulated CD45, increased their granularity (side scatter area - SSC-A) and progressively downregulated CX3CR1, F4/80 and CD68 over the course of infection. At dpi 5 and 6 microglia showed peak expression of several cell surface markers, including CD64, MerTK, CD86, MHC-II, CCR2 (low levels) and Ly6C, which were subsequently downregulated by dpi 7 (Fig. 7c-e). While P2RY12 was expressed by all microglial subsets, average expression was reduced on the total microglial population by dpi 7. This was due to a combination of P2RY12 down-regulation only on P2RY12lo cells and an increase in the proportion of this subset over the course of infection.
In the 4 individual microglial cell phenotypes identified during infection (Fig. 4f, 7e and Additional file 9), the P2RY12hi cells showed a higher expression of all measured markers, compared to P2RY12lo cells. While all phenotypes upregulated CD45 and CD64, the P2RY12hi cells only upregulated CD11b, P2RY12, and CD11c. In contrast, P2RY12lo cells downregulated P2RY12, as mentioned above, and showed no change in CD11b or CD11c expression. Of note, the P2RY12hiCD86− cells upregulated CD86 during infection, becoming P2RY12hiCD86lo microglia (Fig. 5i, S5 and S9).
Changes in microglial immunophenotypes correlated with monocyte infiltration and neuronal infection from dpi 4. Microglia also exhibited a reactive morphology by dpi 5, with hypertrophied cell somata and shortened cytoplasmic extensions (Fig. 7a, b). This was similar in other brain regions, irrespective of the presence of virus (data not shown).
Microglia proliferate early and die late in infection
Using Gating strategy 4, kinetic analysis revealed a decrease in the number of total microglia later in infection, with their proportions reducing dramatically due to the increasing numbers of other leukocytes immigrating into the brain (Fig. 8a, b). Within this population, the number and proportion of the P2RY12hiCD86− microglial subset decreased against an increase in the number and proportion of the other subsets from dpi 4–5 (Fig. 8c, d). Furthermore, notwithstanding the lack of significant change in microglial cell numbers early in infection, microglia proliferated from dpi 4, as shown by the incorporation of BrdU (Fig. 8e-i). At dpi 5, the peak of microglial cell proliferation, P2RY12hiCD86− microglia showed the greatest incorporation of BrdU (Fig. 8g). From dpi 5–7 microglial proliferation decreased (Fig. 8e-i), while the frequency of lymphocyte proliferation increased over this time, and MDM proliferation was minimal (Fig. 8i). Increased myeloid cell numbers in WNV-infected brains, was previously attributed to the infiltration of BM-derived monocytes (Getts 2008, Garber 2019), however, our analysis shows both clear microglial cell proliferation and MDM infiltration.
Coinciding with reduced microglial cell numbers by dpi 7, there was also an increased number and percentage of dead microglia (Annexin V+, Live/Dead stain+) from dpi 6 onwards. However, the proportions of apoptotic microglia (Annexin V+ only) were not significantly different over this time (Fig. 8j-l). Strikingly, the population with the highest proportion of dead and dying cells at dpi 5, 6 and 7 were microglia-like macrophages (Fig. 8l). The increased number of dead microglia explains, at least in part, why proliferation of microglia at dpi 5 did not correspond to an increase in microglial numbers at dpi 6 or 7.
Microglia are the principal producers of IL-12 during lethal WNE
To elucidate the function and contribution of microglia to protective or pathogenic responses in WNE, we stained for a series of intracellular cytokines readily detected without in-vitro stimulation, to minimise non-physiological conditions (Fig. 9). We formerly showed that IL-12, TNF, IFN-g, CCL2, IL-10, IL-1a/b and IL6 are upregulated in WNV-infected brains (1–3, 50). Consistently with previously published work, Ly6Chi macrophages and T cells were the principal source of NO and IFN-g, respectively (Fig. 9a) (3, 50). Ly6Chi macrophages also had the highest expression of CD206, confirming that macrophages can express both pro- and anti-inflammatory markers simultaneously. Microglia-like macrophages expressed NO and CD206 only marginally, suggesting a less inflammatory, alternative role for these cells. Of interest, was the primary production of IL-12/IL-23 p40 by microglia in the later phase of disease (Fig. 9a, d-i). Since IL-23 shares the p40 subunit with IL-12, we performed an ELISA (Fig. 9b) and RNAse protection assay (Fig. 9c) on total brain protein and RNA, respectively, to discriminate between these cytokines. Marginal to no IL-23 (p19/p40) protein or IL-23 (p35) RNA was found in WNV-infected brains, indicating that IL-12 and not IL-23 was most likely to be produced by microglia. Notably, while the P2RY12hiCD86+ microglia subset had the highest frequency of IL-12/IL-23 p40+ cells (Fig. 9i), P2RY12hiCD86− microglia, as the largest subset, produced most of the IL-12 (Fig. 9h). This suggests that these cells have role in T cell activation.