Microglia promote anti-tumour immunity and suppress breast cancer brain metastasis

Abstract

Breast cancer brain metastasis (BCBM) is a lethal disease with no effective treatments. Prior work has shown that brain cancers and metastases are densely infiltrated with anti-inflammatory, protumourigenic tumour-associated macrophages, but the role of brain-resident microglia remains controversial because they are challenging to discriminate from other tumour-associated macrophages. Using single-cell RNA sequencing, genetic and humanized mouse models, we specifically identify microglia and find that they play a distinct pro-inflammatory and tumour-suppressive role in BCBM. Animals lacking microglia show increased metastasis, decreased survival and reduced natural killer and T cell responses, showing that microglia are critical to promote anti-tumour immunity to suppress BCBM. We find that the pro-inflammatory response is conserved in human microglia, and markers of their response are associated with better prognosis in patients with BCBM. These findings establish an important role for microglia in anti-tumour immunity and highlight them as a potential immunotherapy target for brain metastasis.

Extended Data Fig. 1 |. Analysis of TAMs and astrocytes in BCBM.

a Whole mount brightfield and fluorescent microscopy images of metastatic brains (Met1-3) used to generate the scRNA-seq dataset described in Fig. 1d. Metastatic lesions are GFP⁺ (green). Results are representative of a single experiment. Scale bar = 50 mm. b IF staining shows IBA1⁺ cells (red) in normal human brain and three resected patient BCBM tumors. Insets show cell morphology, exhibiting evenly spaced, ramified microglia in normal human brain contrasting heavily infiltrated ameboid microglia in BCBM patients. Results are representative of a single experiment. Scale bar = 50 μm. c Representative FACS plots show gating for single, live (Sytox negative) myeloid cells (CD45⁺CD11b⁺), astrocytes (CD45⁻ASCA2⁺) and 231BR cells (CD45⁻ GFP⁺) isolated for scRNA-seq. d Identification of mouse and human cells by the frequency of reads that align to the mm10 mouse genome. Cutoffs used to identify mouse cells (>0.875 aligned, n = 51,418 cells), human cells (<0.05 aligned, n = 7336 cells) and doublets (0.05–0.875 aligned, n = 913 cells) are shown. e Violin plots show cell distributions for key quality control metrics pre- and post-filtering and removal of poor quality cells. Cells were removed that displayed <500 or >2000 genes (nFeature_RNA), or >10% of genes mapped to the mitochondrial genome (percent mito genes). f Bar chart shows the frequency of cells contributed by each mouse that localize to each cell type in Fig. 1f. g tSNE plot shows astrocytes colored by control or metastatic condition. h Volcano plot shows genes differentially expressed (n = 6,542) between astrocytes from control and metastatic brains determined by Wilcoxon rank sum test, (p < 0.01). See Supplementary Table 2 for full list. Select genes with an absolute value average natural logFC >0.35 are colored and labeled. The y-axis represents the –log10 of Bonferroni corrected Pvalues, and the x-axis represents average natural logFC between conditions.

Extended Data Fig. 2 |. Identification of myeloid cell types and subclustering analysis of proinflammatory microglia.

a tSNE plot shows myeloid cells (n = 15,288) colored and labeled by cell type. mDC = mature dendritic cell. Mono.Macro = monocytes and macrophages. b Dot plot shows top marker genes for each cell type ranked by average natural logFC. Dot size represents the percentage of cells that express the gene, and dot greyscale represents the average expression level. See Supplementary Table 4. c Bar chart shows the frequency of cells contributed by each mouse that localize to each cell type in b. d Feature plots show myeloid cells colored by canonical cell type marker genes or features. Stressed cells were identified by increased expression of mitochondrial genome (percent.mito) genes, and decreased number of genes detected (nFeature_RNA). e Bayesian information criterion (BIC) for microglia topic models from Fig. 2d with the listed number of topics (K), each fit to an error tol = 10. f Bar plot shows the relative enrichment of each topic in control and metastatic animals from Fig. 2d. The relative enrichment was determined by subtracting the average topic assignment for the control mice from the average topic assignment across all cells in each mouse. Highlighted topics show four core topics where all three metastatic mice have a higher relative enrichment than all three control mice (that is min(Metastatic) > max(Control)).

Extended Data Fig. 3 |. Pro-inflammatory marker expression in microglia from BCBM models.

a Gating strategy for identification of microglia. Dot plots (top) show gating for single, live (zombie negative) CD45^loCD11b⁺Ly6C⁻ microglia. Histogram plots (bottom) show subsequent gating for CD74, BST2, and MHC-II in microglia. b Flow cytometry analysis of CD74, BST2 and MHC-II in microglia harvested 14 days post intracardiac injection of 4T1-GFP (100,000) cells into BALB/c animals. Bar graph shows the percent of microglia that express each marker in control (n = 7) and metastatic (n = 7) brains. Pvalues were generated by an unpaired two-sided student’s t-test, and error bars indicate mean +/− standard deviation. c Quantification of microglia in tumor and distal regions of mice bearing EO771-GFP tumors. Representative images (left panels) show microglia localization relative to other cell types using a machine learning classifier (see Methods). Pie graphs (right panels) show the proportion of microglia and other cell types in each region. Frequencies are as follows: other non-microglia cells (TMEM119⁻CD74⁻MHC-II⁻ISG15⁻), distal = 0.94, tumor = 0.22; microglia (TMEM119⁺), distal = 0.05, tumor = 0.22; tumor cells (ISG15⁺TMEM119⁻), distal < 0.01 tumor = 0.14; other immune cells (TMEM119⁻CD74⁺MHC-II⁺), distal < 0.01, tumor =0.41. Scale bar = 100um.

Extended Data Fig. 4 |. Quantification of tumor size in FIRE-WT and FIRE-KO animals.

a IVIS images show EO771 luciferase luminescence signal change over time in FIRE-WT and FIRE-KO animals. Representative animals that displayed continuous signal increase (tumor growth, solid line) vs. signal decrease (tumor rejection, dashed line) are shown. Pseudocoloring of luminescence shows quantification of radiance (p/sec/cm²/sr). b Line graphs show quantification of luminescence signal change over time in all FIRE-WT and FIRE-KO animals. Solid lines indicate animals that demonstrated tumor growth and dashed lines indicate those that showed tumor rejection. Growth was defined by signal increase over time, and rejection was defined as either baseline signal (<10⁶) or >5-fold decrease in signal relative to maximum. c Serial dilution analysis of EO771 cell engraftment in FIRE-WT and FIRE-KO animals. 10–200 × 10⁴ EO771 cells were transplanted intracranially into each mouse strain. Ex vivo whole brain luminescence images show signal from tumor cells in each tissue at day 14. Fractions denote the number of grafts that produced macroscopic tumors in each condition. d Dot plots quantify luminescent signal (total flux) from each tissue shown in Extended Data Fig. 4c at day 14. Pvalues were generated by an unpaired two-sided student’s t-test.

Extended Data Fig. 5 |. Analysis of NK, T, and monocyte responses to BCBM in FIRE-WT and FIRE-KO animals.

a Quantification of tumor burden in FIRE-WT and FIRE-KO animals (n = 8/group). Mice were injected with EO771 GFP-Luc cells as described in Fig. 5a and tumors were harvested and analyzed by IVIS on day 7. Images (left panels) show pseudocoloring of radiance (p/sec/cm²/sr), and bar graph shows quantification of total flux (p/s). Pvalue shown is the result of a student’s unpaired two sided t-test. Error bars represent mean +/− standard deviation. b Box plots show frequency of T cell subsets from Fig. 5c (n = 7 FIRE-WT, n = 8 FIRE-KO). Frequencies shown are out of all T cells. Bounds of box and whiskers are indicative of the first through fourth interquartile range. Pvalue shown is the result of a student’s un-paired two sided t-test. c Analysis of T cell activation in tumor bearing FIRE-WT and FIRE-KO brain tissues by flow cytometry. CD44 and CD62L expressions were measured in CD4 and CD8 T cells to delineate T effector (Teff), T central memory (TCM) and naive T cell subsets. Representative FACS plots (top panels) show gating for each subset after gating for single, live (Sytox negative) cells. Bar graphs (bottom panels) show quantification of T cell counts for each group. Error bars represent mean +/− standard deviation. Pairwise comparisons of counts between groups were not significant. d Quantification of monocytes in tumor bearing FIRE-WT and FIRE-KO brain tissues by flow cytometry. CD11b⁺Ly6C⁺ monocytes were identified following gating for CD3⁻NK1.1⁻, single, live (Sytox negative) cells. Top panels show representative FACS plots, and bottom panels show quantification of cell counts. Error bars represent mean +/− standard deviation. Pairwise comparisons of counts between groups were not significant. e Linear regression model of CD8+ T cell and Treg quantification from Extended Data Fig. 5b. R-squared and Pvalues determined by simple linear regression function.

Extended Data Fig. 6 |. Analysis of tumor burden and the immune response in T cell deficient mice.

a Gating scheme for analysis of T cells in brain tissue harvested from vehicle treated, FTY720 treated and RAG1-K0 mice. FACS plots show gating of TCRb⁺ T cells from single, live (sytox negative) CD45^hiCD11b⁻ cells from Extended Data Fig. 6b. b Gating scheme for analysis of microglia and monocytes in brain tissue from vehicle, FTY720 and RAG1-K0 animals. FACS plots (top panels) show gating for CD45^hi-int and CD11b⁺cell populations, followed by gating for CD45^hiLy6C⁺ monocytes and CD45^intLy6C⁻ microglia (bottom panels). c Bar graphs show the percentage of microglia and monocytes out of total live, single cells in brains harvested from vehicle (veh, n = 6), FTY720 (FTY, n = 6) and RAG1-KO (RAG1, n=4) animals. Pairwise comparisons of percentages between groups were not significant. Error bars represent standard deviation. d Quantification of E0771 tumor burden at endpoint on day 12 by IVIS. Pseudocolor shows radiance (p/sec/cm²/sr) in each whole brain.

Extended Data Fig. 7 |. Analysis of BCBM immune repertoire in T cell deficient and replete mice.

a Dot plot shows top marker genes for each cell type in total CD45^hi-int sorted cells, ranked by the average log₂ fold-change and determined by the Wilcoxon rank sum test. Dot size represents the percentage of cells that express each gene, and dot greyscale represents the average expression level. Macro.DCs = macrophages and dendritic cells, mDCs = mature dendritic cells, pvMacro = perivascular macrophages. Bar graphs illustrate relative contribution of each cluster to total leukocytes, separated by mouse strain and timepoint. b UMAPs show T cells (n = 1949 cells) from C57BL/6 mice at day 4 and 10, colored by cluster (left) and timepoint (right). c Bar graph illustrates the distribution of T cell clusters in each animal (n = 6) separated by timepoint. d Dot plot shows expression of top marker genes for each T cell cluster from Extended Data Fig. 7b. CD4.eff = CD4+ effector T cell, CD8.eff = CD8+ effector T cell, Lt.stg.eff = late stage effector T cell, γδ = gamma delta T cell. e Dot plot shows expression of top marker genes for each microglia cell cluster from Fig. 7b.

Extended Data Fig. 8 |. Analysis of humanized mouse model of BCBM and patient BCBM data.

a Whole mount brightfield and fluorescence microscopy images show brains from MITRG mice transplanted with GFP-labeled iHPSC cells and mCherry-labeled 231BR cells from Fig. 8a. b Identification of mouse and human cells by the frequency of reads that align to the mm10 mouse genome. Cutoffs used to identify mouse cells (>0.95 aligned, n = 641 cells), human cells (<0.1 aligned, n=25,287 cells) and doublets (0.1–0.95 aligned, n = 387 cells) are shown. c Violin plots show cell distributions for key quality control metrics pre- and post-filtering and removal of poor quality cells. Cells were removed that displayed >20% of genes mapped to the mitochondrial genome (percent mito genes). d tSNE plot shows human cells, colored by cluster and labeled by cell type. pvMacro = perivascular macrophages. Cycling = cycling myeloid cells. See Supplementary Table 6 for full gene list. e Dot plot shows top marker genes for each cell type determined by the Wilcoxon rank sum test and ranked by average natural logFC. Dot size represents the percentage of cells that express the gene, and dot greyscale represents the average expression level. See Supplementary Table 6 for full gene list. pvMacro = perivascular macrophages. f Bar chart shows the frequency of cells contributed by each mouse to the cell types shown in e. g tSNE plots colored to show the expression of BST2, CD74 and CCL3. h tSNE plot shows the distribution in human microglia of the three core topics identified in mouse microglia in response to BCBM. Gene scores for each topic from Fig. 2e were generated using the AddModuleScore function in Seurat (See Methods). Each topic is indicated by color, where only cells with a topic score > 0.25 are colored. Contrast gray scale indicates topic weights. Scaling was performed by dividing all topic scores by the maximum topic score across the dataset. i Bar plot shows the relative enrichment of each topic score in human microglia. j Kaplan-Meier plots show overall survival probability stratified by MHC-II, CSF1, and BST2 expression in bulk RNA-seq data from human patient BCBM tumors (Varešlija et al.).

Fig. 1 |. Single-cell analysis of TAMs in BCBM.

a, Schematic showing disease progression in mouse 231BR-Foxn1^nu/nu BCBM experimental metastasis model. A total of 500,000-GFP-Luc-labelled 231BR cells were injected into the left cardiac ventricle of Foxn1^nu/nu mice and collected 28 days later. Whole-mount brightfield and fluorescence microscopy images show a representative brain with GFP⁺ metastatic foci (green). b, IF staining showing IBA1⁺ cells (red) in control and metastatic brains at 7,14 and 28 days post 231BR cell injection. Metastatic cells are GFP⁺ (green). 231BR cells arrest in blood vessels and cross into the brain 2–7 days post-injection, then grow along blood vessels forming micrometastases by day 14, parenchymal metastases by day 28. Scale bar, 50 μm. c, Quantification of IBA1⁺ cells in control (n = 4) and metastatic (n = 4) brains 28 days post 231BR cell injection. IBA1⁺ cells were quantified in control (n = 61 fields) and metastatic tumour regions (n = 41 fields). Bargraph shows 1.95-fold increase of IBA1⁺ cells in tumour compared with control tissue. Pvalue was generated using a two-sided, unpaired t-test, and error bars show mean ± standard deviation. d, Schematic showing experimental design for generation of scRNA-seq dataset. Foxn1^nu/nu mice were injected with 500,000 GFP-Luc-labelled 231BR cells, and brains were collected 28 days later. Three metastatic (Met1-3) and three control (Con1-3) brains were digested and myeloid cells (CD45⁺, CD11b⁺), astrocytes (CD45⁻, ASCA2⁺) and 231BR (CD45⁻, GFP⁺) cells were isolated by flow cytometry for droplet-based scRNA-seq. e, tSNE plot showing mouse cells that passed quality filtering (n = 42,891), coloured and labelled by cell type. f, Dot plot showing top marker genes for each cell type ranked by the average natural log fold change (FC) and determined by the Wilcoxon rank-sum test. Dot size represents the percentage of cells that express each gene, and dot greyscale represents the average expression level. For full marker gene list, see Supplementary Table 1.

Fig. 2 |. Microglia display robust pro-inflammatory responses to BCBM.

a, tSNE plot showing clustering of myeloid cells (n = 24,348), coloured by mouse. b, tSNE plot showing each myeloid cell coloured by MG score, the core microglia gene signature from Bowman et al. (2016) that compared microglia with bone marrow-derived cells using bulk RNA-seq from lineage-labelled mice. For full MG score gene list, see Supplementary Table 3. Scores were calculated using the AddModuleScore function in Seurat. Top marker genes (grey) for each myeloid cell type were identified using the Wilcoxon rank-sum test. For myeloid cell type markers, see Supplementary Table 4. mDC, mature dendritic cell; Mono/Macro, monocytes and macrophages, c. Bar plot showing selected top GO terms associated with the BCBM response microglia signature. This signature was generated by differential gene expression analysis of microglia from metastatic versus control brains (n = 632 upregulated genes, adjusted P < 0.05). Differentially expressed (DE) genes were determined using the Wilcoxon rank-sum test. GO terms were identified using MouseMine, and select upregulated terms with Holm-Bonferroni-adjusted Pvalues < 0.05 were retained. For DE genes, see Supplementary Table 4. d, Schematic overview of topic model fitting method to assess microglia heterogeneity. The CountClust R package was used to fit a topic model using LDA. A matrix for ‘gene weights’ was generated that contains a list of the genes comprising each topic and the gene weight (Supplementary Table 5). A second matrix for ‘topic weights’ lists the weight of each topic across the cells. e, Heatmap showing three core topics upregulated in microglia in BCBM. Scaled gene weights for top genes comprising each topic are shown. f, tSNE plots showing distribution of three core topics in each microglia. Left panels show topic weight in each cell indicated by contrast greyscale. Right panel overlay shows top topic assignment for each cell, where only cells with a topic weight >0.1 are coloured. g, tSNE plots showing expression of selected genes from each topic in myeloid cells.

Fig. 3 |. The microglia pro-inflammatory response is conserved in diverse BCBM models.

a, Flow cytometry analysis of CD74, BST2 and MHC-II expression in microglia from four BCBM models. Microglia were identified by gating on CD45^lo, CD11b⁺, Ly6⁻ C cells^,. Bar graphs show the percentage of microglia expressing each marker. Pvalues were generated by an unpaired two-sided Student’s t-test, error bars indicate mean ± standard deviation. b, In situ analysis of microglia pro-inflammatory marker expression by multiplex IF (CODEX). Brain tissue slices from mice bearing E0771-GFP tumours were stained for DAPI (blue), and antibodies against GFP (green), TMEM119 (red), MHC-II (white), CD74 (yellow) and ISG15 (cyan). Left: overview of all markers. Scale bar, 840 μm. Right: pairwise marker expression in higher-magnification insets of tumour and distal regions. Short arrows indicate representative microglia expressing AP markers (TMEM119⁺ MHC-II⁺CD74⁺), and long arrows indicate representative microglia expressing IFN response marker (TMEM119⁺1SG15⁺). Results are representative of two independent experiments. Scale bar, 100 μm. c, Quantification of pro-inflammatory markers in brain tissue slices. Microglia were identified on the basis of TMEM119 expression and then scored for marker expression. Left: images showing phenotype in representative tumour and distal regions. Tumour cells (ISG15⁺TMEM119⁻) and other non-microglia cells (TMEM119⁻CD74⁻MHC-II⁻SG15⁻) are shown in green and grey, respectively. Right: pie graphs showing the proportion of microglia displaying marker combinations. Frequencies are as follows: CD74⁺MHCII⁺ISG15⁺, distal <0.01, tumour 0.29; CD74⁺MHC-II⁺ISG15⁻, distal <0.01, tumour 0.11; CD74⁻MHC-II⁺-SG15⁺, distal <0.01, tumour 0.07; CD74⁺MHC-II⁺ISG15⁺, distal <0.01, tumour 0.07; CD74⁺MHC-II⁻ISG15⁻ distal 0.01, tumour 0.08; CD74⁻MHC-II⁺ISG15⁻, distal <0.01, tumour 0.02; CD74⁻MHC-II⁺ISG15⁺, distal 0.01, tumour 0.11; CD74⁻MHC⁻II⁻ISG15⁺, distal 0.97, tumour 0.24. n = 4 mice per condition. Scale bar, 100 μm. d, Analysis of cytokine expression by microglia in BCBM. Microglia were isolated from control (n = 4) and metastatic (E0771-C57BL/6, n = 8) brains by flow cytometry, and cell lysates were analysed by cytokine array (Eve Technologies). Pvalues shown are the result of a two-sided unpaired Mann-Whitney t-test.

Fig. 4 |. Animals lacking microglia demonstrate reduced capacity for tumour rejection.

a, Schematic depiction of Csf1r^{ΔFIRE/ΔFIRE} mouse model. Deletion of FIRE super-enhancer in FIRE-KO mice leads to loss of CSF1R protein expression in specific tissues. In the CNS, microglia do not develop, while monocyte and macrophage numbers are unaffected. b, Representative flow cytometry plots show the percentage of CD45^loCD11b⁺ microglia and CD45^hi immune cells gated from live (Sytox negative), single cells in FIRE-WT (n = 2) and FIRE-KO (n = 2) mouse brains. c, Schematic of experimental design to compare disease progression in FIRE-WT and FIRE-KO mice. FIRE-WT (n = 19) and FIRE-KO (n = 14) mice were injected intracranially with 100,000 GFP- and luciferase (GFP-Luc)- labelled E0771 cells. Control FIRE-WT mice (n = 8) were also injected with PBS. Animals were imaged for luminescence (IVIS) every 3 days before dissection at endpoint on day 14. d, Kaplan-Meier plot shows survival in FIRE-WT (19/19,100%) and FIRE-KO (9/14,64%) mice. Pvalue determined by log-rank (Mantel-Cox) test. e, Bar graph shows percentage body weight change for surviving PBS-injected (n = 8), FIRE-WT (n = 19) and FIRE-KO (n = 9) animal from d at day 14 relative to day 0. Pvalues determined by unpaired two-sided Student’s t-test, and error bars represent mean ± standard deviation. f, Bar graph summarizing the frequency of animals that displayed tumour growth and tumour rejection in FIRE-WT and FIRE-KO mice. Tumour rejection was defined by a lack of engraftment or engraftment followed by tumour rejection. Pvalue was determined by two-sided Fisher’s exact test.

Fig. 5 |. Microglia promote NK and T cell responses to BCBM.

a, Schematic of experimental design to compare NK and T cell responses in FIRE-WT and FIRE-KO E0771 tumour-bearing mice. FIRE-WT (n = 8) and FIRE-KO (n = 8) mice were injected intracranially with 100,000 E0771 GFP-Luc cells. Animals were imaged for luminescence (IVIS) on days 1,4 and 6before dissection on day 7. b, Analysis of NK and T cell subsets in FIRE-WT (n = 7) and FIRE-KO (n = 8) mice by flow cytometry. Representative FACS plots show gating for each NK and T cell subset after gating for single, live (Sytox-negative) cells. c, Bar graphs showing cell counts for NK and T cell subsets. Counts shown are out of 100,000 single, live cells analysed. Pvalues are the result of a Student’s unpaired two-sided t-test. Error bars represent mean ± standard deviation. d, Analysis of CD107a expression in NK and CD8⁺ T cells by flow cytometry. FACS plots (left) show expression of CD107a from spleen and brains of representative animals from each cohort. Bar graph shows cell counts out of 100,000 single, live cells. Pvalues are the result of a Student’s unpaired two-sided t-test. Error bars represent mean ± standard deviation.

Fig. 6 |. Microglia and T cells coordinate the anti-tumour response.

a, Schematic of experimental design to determine effects of T cell deficiency on BCBM. Tumour burden was compared in three cohorts of animals, FIRE-WT vehicle treated (Veh., n = 13), FIRE-WT FTY720 treated (FTY, n = 6) and RAG1-KO (RAG1, n = 12). Vehicle (PBS + 0.1% DMSO) or FTY720 (5 mg kg⁻¹) were administered via intraperitoneal injection to FIRE-WT animals on day 0 and repeated daily. A total of 70,000 E0771 GFP-Luc cells were delivered to each animal in all three cohorts by intracranial injection on day 0 following drug delivery. Brain tissues were collected at endpoint on day 12 and analysed for tumour burden by IVIS and immune response by flow cytometry. b, Bar graph showing the percentage of TCRb⁺ T cells in brain tissues collected from each cohort (n = 6 Veh., n = 6 FTY, n = 4 RAG1) of animals at endpoint, gated out of single, live (Sytox negative) CD45^hl cells as shown in Extended Data Fig. 6a,b. Pvalues shown are the result of an unpaired two-sided Student’s t-test. Error bars represent mean ± standard deviation. c, Quantification of E0771 tumour engraftment at endpoint on day 12 by IVIS. Bar graph shows frequency of animals in vehicle (Veh.), FTY720 (FTY) and RAG1-KO (RAG1) groups that grew tumours. P = 0.51 Veh. versus FTY. and P = 0.21 Veh versus RAG1 by two-sided Fisher’s exact test. d, Quantification of E0771 tumour burden at endpoint on day 12 by IVIS. Box-and-whisker plots show total flux per brain of vehicle (Veh.)- and FTY720 (FTY)-injected FIRE-WT and RAG1-KO (RAG1) cohorts. Bounds of box and whiskers are indicative of the first through fourth interquartile range. Pvalues shown are the result of an unpaired two-sided Student’s t-test. e, Analysis of pro-inflammatory marker expression in microglia from T cell-deficient mice (n = 6 Veh., n = 6 FTY, n = 4 RAG1). Left: FACS plots showing expression of MHC-II, CD74 and BST2 in representative animals, following gating on single, live (Sytox-negative) CD4S^intCD11b⁺Ly6c^neg microglia as shown in Extended Data Fig. 6b. Right: bar graphs showing the percentage of marker positive microglia in each cohort. Pvalues are the result of an unpaired two-sided Student’s t-test. Error bars represent mean ± standard deviation.

Fig. 7 |. Altered microglia activation in animals lacking T cells.

a, Schematic of experimental design to evaluate changes in microglia and T cell activation over time. A total of 100,000 EO771 GFP-Luc cells were administered through intracranial injection to C57BL/6 (n = 6) and RAG1-KO (n = 6) mice at day 0. Brain tissues were collected 4 days (n = 3 per group) and 10 days (n = 3 per group) post injection and sorted for live, CD45^hi-int cells by flow cytometry for scRNA-seq analysis. b, UMAPs showing all immune cells (n = 31,053 cells) (left), microglia coloured by subcluster (middle) and microglia coloured by condition (right). UMAPs for microglia were downsampled to display an equal number of microglia from each condition (n = 1,000 cells per condition). c, Bar graph showing the mean topic score for each programme (AP, secretory, and IFN response) in all microglia from each condition. d, Violin plots quantifying the expression of key markers of the secretory, IFN response and AP programmes in microglia from each condition. e, Feature plots illustrating the distribution of key markers of the secretory, IFN response and AP programmes in microglia. f, Heat map of log₂ fold change (FC) of key markers of the secretory (Secr, top), IFN response (IFNR, middle) and AP (bottom) programmes separated by timepoint and mouse strain. g, UMAP plot of pseudotemporal cell ordering results performed using Monocle 3 showing microglia cell state ordering beginning with the homeostatic state (pseudotime 0). Violin plot shows the contribution of each microglia cluster at specific pseudotime values. Microglia cell states are ordered by the median pseudotime value displayed as a black bar.

Fig. 8 |. The pro-inflammatory response is conserved in human microglia and associated with better prognosis in patients with BCBM.

a, Schematic showing experimental design for scRNA-seq of human microglia from humanized MITRG mice transplanted with 231BR cells. MITRG mouse pups were injected with GFP-labelled iHPSCs, aged to 10 weeks and injected intracardiac with mCherry-labelled 231BR cells. Brains from control (n = 3) and metastatic (n = 3) mice were digested to single-cell suspensions 3 weeks later. Dissociated cells from each sample were indexed using MULTI-seq. Mouse cells were removed using anti-mouse MHC-I magnetic beads, and recovered cells were pooled into metastatic or control samples for scRNA-seq. b, tSNE plot shows human cells (n = 21,353) coloured by mouse and labelled by cell type. Top marker genes (grey) for each cell type were identified using the Wilcoxon rank-sum test. For full marker gene list, see Supplementary Table 6. pvMacro, perivascular macrophages. c, Bar plot showing selected top GO terms associated with the human BCBM microglia response signature. Differentially expressed (DE) genes (n = 4,904, adjusted P< 0.05) were determined using the Wilcoxon rank-sum test. GO terms were determined using Enrichr and select upregulated terms with Pvalues <0.05 were retained. For full gene list, see Supplementary Table 6. d, Kaplan-Meier plot showing overall survival probability in human patients with BCBM stratified by expression of canonical microglia genes. Bulk RNA-seq data from patient BCBM tumours (n = 20, Varešlija et al.) were scored for microglia gene signature and stratified into high and low groups. Scores were determined using the sum of scaled and centred values from log(CPM +1) transformed data, HR=hazard ratio. e, Model for role of microglia in promoting anti-tumour immunity. In microglia-replete conditions (+Microglia), microglia respond to BCBM by upregulating pro-inflammatory programmes (IFN response, AP and secretory) that promote anti-tumour CD4, CD8 and NK cell responses and tumour regression in the CNS. In microglia-depleted conditions (-Microglia), NK and T cell responses are deficient and the proportion of Tregs is increased, resulting in tumour progression. In animals lackingT cells (-T cells), microglia fail to upregulate AP genes and tumour regression is not observed, suggesting that T cells are required for complete microglia activation and that reciprocal microglia-T cell activation is critical for tumour suppression.