From June 2018 to March 2019, we enrolled patients with WHO grade III AAmut or AOD who underwent craniotomy surgery at Beijing Tiantan Hospital (Beijing, China), and blood and tumor tissues were obtained. All these patients were diagnosed and confirmed by histopathological and molecular analysis. Before sampling, none of these enrolled patients used glucocorticoids. The current study was approved by the Institutional Review Board and Ethics Committee of Beijing Tiantan Hospital, Capital Medical University. Written informed consent was obtained from all patients.

After the operation, the ice-cold Dulbecco’s phosphate-buffered saline (DPBS, Sigma-Aldrich) was immediately used to wash AAmut or AOD tumor tissues. In brief, type IV collagenase (Gibco) was used to dissociate the AG specimens. Next, Dulbecco’s modified Eagle’s medium (DMEM, Sigma-Aldrich) was used to wash the specimens. After centrifugation, the specimens were filtered through a 40 µm cell strainer with DPBS and washed with red blood cell (RBC) lysis buffer (BD Biosciences). Next, DPBS was used to wash the dissociated cell suspension. Finally, the cells were resuspended in staining buffer (DPBS containing 5% fetal bovine serum; ScienCell).

Peripheral blood specimens were gathered with ethylenediaminetetraacetic acid anticoagulation tubes. To remove plasma, the blood specimens were centrifuged first. Then, the blood specimens were transferred into SepMate peripheral blood mononuclear cell (PBMC) isolation tubes containing Ficoll (STEMCELL Technologies). After centrifugation, RBC lysis buffer was used to wash the cells. Finally, these cells were washed with staining buffer twice.

A panel of 33 antibodies was used as previously reported (10). Preconjugated antibodies were purchased from Fluidigm Company. Purified antibodies were purchased from Biolegend Company and then conjugated with metals using the Maxpar^® X8 Multimetal Labeling Kit (Fluidigm) according to the manufacturer’s protocol. Supplementary Figure S1 demonstrates the list of the antibodies and reporter isotopes. In brief, the cell specimens were rewarmed quickly. Anti-CD45 antibody conjugated with 156Gd was used to stain cells from AG tissues, while anti-CD45 antibody conjugated with 89Y was used to stain cells from PBMCs. We mixed together the cells from the AG and PBMC samples and then stained the specimens with cell surface antibodies. Subsequently, the mixed specimens were permeabilized and stained with intracellular antibodies. Then 0.125 nM Intercalator-Ir (Fluidigm) diluted in phosphate-buffered saline (PBS; Sigma-Aldrich) containing 2% formaldehyde was used to wash and incubate the antibody-labeled specimens. Specimens were stored at 4°C until CyTOF examination. Before acquisition, deionized water was used to wash the specimens. The specimens were resuspended in deionized water containing a 1:20 dilution of EQ Four Element Beads (Fluidigm) at a concentration of 1 × 10⁶ cells/ml. The specimens were examined by CyTOF2 mass cytometry (Fluidigm Company).

The.fcs files of CyTOF data were uploaded and analyzed with Cytobank (www.cytobank.org). As previously described (11), based on EQ Four Element Beads, we can use the MATLAB-based normalization technique according to the bead intensities. T cells were characterized as CD45+CD3+; natural killer (NK) cells were characterized as CD45+CD3−CD16+CD56+ (8, 12); B cells were characterized as CD45+CD19+; monocytes were characterized as CD45+CD14+CD16+ (13); macrophages or microglial cells were characterized as CD45+CD11b+CD3−CD19− CD66b−CD16− (14); regulatory t cells (Tregs) were characterized as CD45+CD4+CD25+CD127− (15) and granulocytes were characterized as CD45+CD66b+. Mononuclear phagocytes are composed of monocytes and macrophages (16). Immunocyte populations of interest were manually gated as previously reported (17). The viSNE analysis of T cells or glioma-associated microglia/macrophages (GAMs) was performed based on patients with more than 500 cell events in both PBMC and AG tumor lesions. Automatic cluster gate functionality was applied for the hierarchical cluster analysis. R software (version 3.4.0) was used to generate the heatmaps of marker expression or relative marker expression.

For Figures 2E, D we used log10-scaled values to normalize the data.

For Figure 3A , we first calculated the ratio of the value of each GAM cytokine or marker to that of the paired mononuclear phagocytes in PBMCs. Then we log10-scaled the ratio to normalize the values.

Three AAmut and three AOD samples were collected for polychromatic immunofluorescence staining. Four percent formalin was used to fix the AG specimens and the specimens were embedded in paraffin blocks. For polychromatic immunofluorescence, 3 µm paraffin sections were washed in PBS twice, and permeabilized in 0.2 to 0.5% Triton X-100 (Solarbio). Then the paraffin sections were blocked in 5% normal donkey serum (Jackson Lab) and stained with primary antibody. Fluorescent-conjugated secondary antibodies (ZSGB-Bio) were used to detect the primary antibodies. Fluorescence mounting medium (Dako) was used to mount the sections. As previously described (18), we used the Opal 4-Color Manual IHC Kit (Perkin Elmer) for the analysis of formalin-fixed paraffin-embedded AG sections following the manufacturer’s protocols. Zeiss LSM880 NLO microscope was used to acquire fluorescent images. Primary antibodies were anti-CD45 (OriGene), anti-lba1 (CST), and anti-CD206 (Proteintech). GAMs were defined based on cells that costained with CD45 and lba1. CD206+ GAMs were defined based on cells that were costained with CD45, lba1, and CD206. The percentage of CD206+ GAMs was defined by (CD206+ GAMs)/GAMs.

For the CyTOF data, five AAmut samples and the paired PBMCs and 10 AOD samples and the paired PBMCs were analyzed. The paired t-test was used to determine significant differences between the AG and paired PBMC samples. The unpaired t-test was used to determine significant differences between the AAmut and AOD lesions. GraphPad Prism software (version 7.00) was used to perform statistical analysis. P values less than 0.05 were considered to be statistically significant.

We obtained 10 AOD tumor tissues, all of which had paired peripheral blood samples (oPBMCs). We also obtained five AAmut tumor tissues and paired PBMC (aPBMC) samples ( Figure 1A ). The baseline characteristics of all AG patients are summarized in Table 1 .

We mapped the immune compartments of the AOD and AAmut lesions and their paired PBMCs at the same time ( Figure 1A ). The initial gating hierarchies for CD45+ immunocytes are demonstrated in Supplementary Figure S2A , and Supplementary Figure S2B summarizes the gating strategies for the indicated immunocytes. The viSNE map of the CD45+ immunocytes collected from all AG specimens demonstrated differential abundances of infiltrating immunocyte populations in the tumor immune milieu compared to those in the PBMCs ( Supplementary Figure S2C ).

We mapped the immune compartment of the AG tumor lesion and the paired peripheral blood specimens at the same time to distinguish the tumor-driven immune changes from the AG immune environment. In the AG immune microenvironment, mononuclear phagocytes (64.16% in AAmut and 56.76% in AOD) and T lymphocytes (25.92% in AAmut and 31.9% in AOD) were the most abundant immunocytes. There were no significant differences in the compartments of immunocytes between the AAmut and AOD immune microenvironments, and the immunocyte compartments in the peripheral blood were also similar. Compared with that in the PBMCs, the ratio of mononuclear phagocytes was notably increased in the AG lesions (p < 0.001 in both AAmuts and AODs), while the ratios of T cells (p < 0.01 in AAmuts and p < 0.001 in AODs) and B cells (p < 0.05 in AAmuts and p < 0.01 in AODs) were notably decreased, and the ratios of NK cells and granulocytes were similar ( Figures 1B, C ).

Compared with those in the PBMCs, the percentages of CD4+ T cells (p < 0.01 in both AODs and AAmuts) declined, while those of CD8+ T cells (p < 0.01 in both AAmuts and AODs) increased in the AAmuts and AODs. As expected, the proportions of Tregs in the AG lesions were significantly increased in both the AAmuts and AODs (p < 0.05 and p < 0.001, respectively). PD-1-, TIM-3-, or LAG-3- positive T cells are recognized as exhausted subgroups (19–21). Compared to that in the PBMCs, the proportions of PD-1- or TIM-3- positive exhausted T cells were substantially higher at the AG tumor sites ( Figure 2A ).

We employed the viSNE map tool (22) to convert the high-dimensional CyTOF data into a two-dimensional atlas. ViSNE analysis was performed on the patients who gathered more than 500 T cells in both the tumor sites and PBMCs. Finally, five AAmut patients and five AOD patients were analyzed. The viSNE map demonstrates that T cells in the AAmut and AOD groups displayed similar distributions ( Figure 2B ). Compared with the PBMCs, the AG lesions had a certain group of T cells that highly expressed PD-1 ( Figure 2C ).

Based on the hierarchical cluster analysis of the T cells using automatic cluster gate functionality, the T cells were subdivided into 21 subgroups according to the surface markers ( Figure 2D ). The heatmap visualized the expression profiles of these T cell clusters ( Figure 2E ). We identified seven CD4+ phenotypes, eleven CD8+ phenotypes, one Treg phenotype, and two CD4+/ CD8+ double-negative phenotypes with this approach.

Previous studies have shown the strong infiltration of peripheral macrophages and resident microglia within gliomas (23), and macrophages and microglia are collectively termed GAMs. In the current research, GAMs were the most enriched immune cell population in the AG tumor sites compared to the other immune cells. These cells were obviously distinguishable from the mononuclear phagocytes in PBMCs; the GAMs in both AAmut and AOD had higher expression levels of PD-L1, IDO, LAG-3, TIM-3, CD206, and TNFα than mononuclear phagocytes in the PBMCs ( Figure 3A ). Moreover, the GAMs showed intertumoral heterogeneity since CD206, immune checkpoints (PD-L1, TIM-3, and LAG-3), immunosuppressive cytokines (IL-10 and TGFβ), TNFα, and VEGF were expressed at various levels in the AG patients ( Figure 3A ). Compared to the mononuclear phagocytes in the PBMCs, GAMs in AOD lesions expressed higher levels of TGFβ (p < 0.05). Although TGFβ is an immunosuppressive agent (24), there was no significant difference in TGFβ expression levels in GAMs between AAmut and ADO lesions ( Supplementary Figure S3A ).

Four AAmut patients and four AOD patients gathered more than 500 GAM or mononuclear phagocyte event counts in both the AG tumor lesions and the PBMCs, and viSNE analysis was performed based on these cells. The viSNE map demonstrated that the GAMs were obviously distinguishable from the mononuclear phagocytes in PBMCs ( Figure 3B ). With automatic cluster gate functionality, the GAMs or mononuclear phagocytes could be subdivided into 17 subgroups based on the surface markers, and the expression levels of these GAM subpopulations were visualized in a heatmap ( Figures 3C, D ). At the single-cell level, the viSNE map demonstrated a cluster involving M-1, which was described as high expression of CD206, a marker that is expressed by protumor GAMs and may promote a tumor-supportive microenvironment (25). This cluster was excluded from the PBMCs, and there were more CD206+ GAMs in the AAmut lesions than in the AOD lesions (p < 0.05) ( Figure 3E and Supplementary Figure S3B ). Polychromatic immunofluorescence confirmed that there were more CD206+ GAMs in the AAmut lesions than in the AOD lesions (p < 0.01) ( Figure 3F and Supplementary Figure S3C ).

It has been reported that CXCR3 is required for NK cell infiltration (26). The expression level of CXCR3 between AAmut and AOD was not significantly different whereas the infiltrated NK cells in the AAmut or AOD lesions expressed higher CXCR3 levels (p < 0.05 or p < 0.001 respectively) than those in their paired PBMCs ( Figure 4 ). Although the difference in IFNγ expression level between NK cells in AAmut and ADO lesions was not significant, the NK cells that infiltrated the AOD lesion seemed to demonstrate higher levels of cytolytic activities, as these NK cells expressed higher levels of IFNγ than the paired PBMCs (p < 0.01). Notably, granzyme B expression levels were similar between AAmut and AOD tumor sites, whereas granzyme B expression level was significantly lower in the AOD samples than in the peripheral blood samples (p < 0.05) ( Figure 4 ).