Abstract

   Triggering receptor expressed on myeloid cells 2 (TREM2) plays
   important roles in brain microglial function in neurodegenerative
   diseases, but the role of TREM2 in the GBM TME has not been examined.
   Here, we found that TREM2 is highly expressed in myeloid subsets,
   including macrophages and microglia in human and mouse GBM tumors and
   that high TREM2 expression correlates with poor prognosis in patients
   with GBM. TREM2 loss of function in human macrophages and mouse myeloid
   cells increased interferon-γ–induced immunoactivation, proinflammatory
   polarization, and tumoricidal capacity. In orthotopic mouse GBM models,
   mice with chronic and acute Trem2 loss of function exhibited decreased
   tumor growth and increased survival. Trem2 inhibition reprogrammed
   myeloid phenotypes and increased programmed cell death protein 1
   (PD-1)^+CD8^+ T cells in the TME. Last, Trem2 deficiency enhanced the
   effectiveness of anti–PD-1 treatment, which may represent a therapeutic
   strategy for patients with GBM.
     __________________________________________________________________

   TREM2 inhibition remodels myeloid cells in the glioblastoma
   microenvironment into a proinflammatory state, curbing tumor growth.

INTRODUCTION

   Glioblastoma (GBM), the most common primary malignant brain tumor,
   continues to be a challenging disease, with short-term survival despite
   aggressive multimodal therapy ([61]1). A major obstacle to effective
   GBM treatment is the “cold” tumor microenvironment (TME). Deficiencies
   in T cell infiltration and active proinflammatory responses render GBM
   resistant to many immunotherapies, including immune checkpoint blockade
   ([62]2). Myeloid cell populations including macrophages and microglia,
   which account for more than 50% of leukocytes enriched in the TME
   ([63]3, [64]4), rarely exert tumoricidal functions and instead maintain
   an immunosuppressive microenvironment. The intrinsic phenotypic
   plasticity of myeloid cells contributes to adaptation to the brain’s
   local inflammatory status and evolution into protumorigenic cell
   populations. Thus, approaches that remodel myeloid cell phenotypes are
   critical for developing new strategies for GBM treatment.

   The myeloid cell receptor, triggering receptor expressed on myeloid
   cells 2 (TREM2), has been studied extensively for its function in
   microglia in neurodegenerative diseases such as Alzheimer’s disease
   ([65]5–[66]7). Recently, studies have revealed that certain solid
   tumors express a high level of TREM2 ([67]8–[68]11), but the role of
   TREM2 in cancer appears to be tumor-type dependent. For instance, TREM2
   suppressed the malignant progression of colorectal cancer and afforded
   protection from hepatocellular carcinoma ([69]8, [70]9). On the
   contrary, increased TREM2 expression in non–small cell lung and gastric
   cancers was associated with poor prognosis ([71]10, [72]11). One
   possible explanation for this inconsistency is that in some tumors,
   TREM2 can be expressed by cancer or stromal cells, which may result in
   differential effects on tumor progression ([73]12). However, there is
   increasing evidence that TREM2 is widely expressed in tumor-associated
   macrophages, ([74]13) and TREM2 loss of function altered the tumor
   macrophage landscape in the TME of solid tumors of the body, leading to
   improved efficacy of immune checkpoint blockade ([75]14, [76]15).
   However, little is known about the role of TREM2 in GBM progression and
   the mechanisms of how TREM2 regulates myeloid cell function ([77]16).

RESULTS

TREM2 expression is higher in GBM tumors than normal brain tissue and
inversely correlates with patient outcome

   To determine the clinical relevance of TREM2 in human GBM, we first
   asked whether TREM2 expression is altered in IDH1/2–wild-type GBM
   tumors in The Cancer Genome Atlas. We found that TREM2 mRNA levels are
   significantly higher in IDH1/2–wild-type GBM tumors compared to normal
   brain tissue ([78]Fig. 1A). We also found that in gliomas in general,
   TREM2 expression positively correlates with grade, with the highest
   levels found in grade 4 gliomas (fig. S1). To ascertain whether TREM2
   levels are associated with clinical outcomes, we stratified
   IDH1/2–wild-type GBM tumors into high and low TREM2 expressers and
   found that high TREM2 expression correlates with shorter median
   progression-free survival (PFS) and overall survival (OS) ([79]Fig. 1,
   B and C). To identify the specific cell type expressing TREM2 in GBM
   tumors, we examined our single-cell RNA sequencing (scRNA-seq) dataset
   using fresh tumor samples from two patients with GBM ([80]17). As
   reported ([81]18), among malignant tumor cells, there were four typical
   cellular states, including astrocyte-, mesenchymal-, neural progenitor
   (NPC)–, and oligodendrocyte progenitor–like states ([82]Fig. 1D). Among
   CD45^+ immune cells, unsupervised clustering by Uniform Manifold
   Approximation and Projection (UMAP) identified seven cell clusters
   corresponding to typical immune cell markers, such as microglia,
   macrophages, T cells, natural killer (NK) cells, and dendritic cells
   ([83]Fig. 1D). Microglia and macrophages occupy the major fractions of
   leukocytes ([84]Fig. 1D). In contrast to tumor cells, we found high
   expression of TREM2 in myeloid cell populations, including microglia,
   macrophages, and myeloid-derived suppressor cells (MDSCs)
   ([85]Fig. 1E). Thus, in GBM, we hypothesized that TREM2-expressing
   tumor-associated myeloid populations might play important roles in
   tumor progression.

Fig. 1. TREM2 expression is higher in GBM tumors than normal brain tissue and
inversely correlates with patient outcome.

   [86]Fig. 1.
   [87]Open in a new tab

   (A) Box plots of TREM2 expression in IDH1/2–wild-type GBM tumors (n =
   252) and normal brain specimens (n = 10). Wilcoxon test, P = 5.1 ×
   10^−7. (B) Kaplan-Meier curves show a PFS of 249 patients with
   IDH1/2–wild-type GBM based on TREM2 expression. TREM2 expression was
   dichotomized into high/low according to the optimal cutoff. Log-rank
   test, P = 0.012. (C) Kaplan-Meier curves show an OS of 115 patients
   with IDH1/2–wild-type GBM by high and low TREM2 expression. Patients
   were divided on the basis of the optimal cutoff of TREM2 expression.
   Log-rank test, P = 0.012. (D) UMAP plots of cell clusters identified in
   human GBM tumors. (E) UMAP and dot plots of TREM2 expression level in
   each cluster. AC, astrocyte; MES, mesenchymal; OPC, oligodendrocyte
   progenitor. *P < 0.05.

Mouse GBM tumors show abundantly infiltrated Trem2^+ myeloid cells

   We explored the role of TREM2 using the IDH1–wild-type syngeneic mouse
   model SB28 ([88]19), which recapitulates major characteristics of human
   GBM, such as resistance to radiotherapy and temozolomide treatment
   ([89]20, [90]21), resistance to immune checkpoint blockade, and high
   fraction of myeloid cells in the TME ([91]19, [92]22, [93]23). Similar
   to human GBM tumors, scRNA-seq analysis of in vivo SB28 tumors revealed
   that macrophage/monocyte-lineage and microglial cells were predominant
   among all immune cell subsets ([94]Figs. 1D and [95]2A). We further
   analyzed Trem2 mRNA levels across all cell clusters and found that
   Trem2 was highly expressed by microglia, M2-like macrophages
   (CD204^+MRC1^+), and nonclassical monocytes (Ly6C^lowCCR2^−), which
   play anti-inflammatory roles in tissue homeostatic regulation
   ([96]Fig. 2B). Relatively low expression was found in M1-like
   interferon-γ (IFNγ)–induced cells and certain B cells ([97]Fig. 2B).
   Notably, the microglial subset in both human and mouse GBM tumors
   expressed the highest levels of TREM2 ([98]Figs. 1E and [99]2B). We
   examined the presence of myeloid cells in SB28 tumor in vivo by
   histopathology ([100]Fig. 2C). An abundance of ionized calcium-binding
   adapter molecule 1 (Iba1)^+ cell populations, which mark both microglia
   and macrophages, was found to accumulate both inside and around tumor
   areas ([101]Fig. 2D).

Fig. 2. Mouse GBM tumor shows abundant Trem2^+ myeloid cells in the TME.

   [102]Fig. 2.
   [103]Open in a new tab

   (A) UMAP plots of cell type clusters in mouse SB28 model of GBM in
   vivo. (B) Dot plots of TREM2 expression level in each cell cluster. (C)
   Representative histology of brain GBM (SB28). Cryosections of 30 μm
   thickness (1 to 6) by hematoxylin and eosin staining (top). Scale bar,
   2 mm. Histological images for invasive tumor front on section 6: Z1,
   Z2, Z3, and Z4. Scale bar, 50 μm. (D) Representative immunostaining
   images. Whole-brain sections (left). Red, Iba1; green, SB28 tumor cells
   (tracking dye); blue, nucleus [4′,6-diamidino-2-phenylindole (DAPI)].
   Scale bar, 2 mm. Iba1^+ myeloid cells around and inside the tumor
   (right). Arrows indicate SB28 cells (yellow) and Iba1^+ cells (white).
   Scale bars, 100 μm.

Trem2 loss of function in vivo inhibits tumor growth and prolongs animal
survival

   We found that there was no detectable Trem2 expression in mouse GBM
   cell line SB28 and an additional syngeneic mouse GBM cell line, NPA
   C54B (fig. S2A) ([104]24). To address the function of Trem2-expressing
   myeloid cells in GBM tumor growth in vivo, we took advantage of Trem2
   transgenic mice ([105]Fig. 3A). Using both SB28 and NPA C54B orthotopic
   GBM models, Trem2^−/− mice showed increased OS and slower tumor growth
   in vivo compared to tumor-bearing Trem2^+/+ mice ([106]Fig. 3B and fig.
   S2B). Moreover, the average weight of tumors dissected from Trem2^−/−
   mice was much lower than Trem2^+/+ mice following orthotopic injection
   of SB28 cells ([107]Fig. 3C). No obvious sexual dimorphism in terms of
   animal survival was observed in SB28-bearing Trem2^−/− mice (fig. S2C).
   There were more apoptotic cells in tumors from Trem2^−/− mice than
   those from Trem2^+/+ mice by cleaved caspase 3 immunohistochemistry
   ([108]Fig. 3D), suggesting that Trem2 deficiency in the TME inhibits
   tumor growth by increasing tumor cell death.

Fig. 3. Trem2 inhibition in vivo inhibits tumor growth and improves animal
survival.

   [109]Fig. 3.
   [110]Open in a new tab

   (A) Trem2^−/− mice genotype verified by real-time quantitative
   polymerase chain reaction (RT-qPCR) (n = 3). Primers’ loci in Trem2
   mRNA were shown. Trem2 mRNA detected in brain tissue from Trem2^+/+,
   Trem2^+/−, and Trem2^−/− (Trem2^−/−_1 and Trem2^−/−_2) littermate mice.
   Control BV2 (scr) and Trem2 knockdown BV2 cells (Trem2i.1 or Trem2i.2)
   served the control. Comparisons performed as indicated. NCBI, National
   Center for Biotechnology Information. (B) Trem2^−/− mice showed
   increased neurological deficit-free survival time than Trem2^+/+ mice
   in both SB28 and NPA C54B GBM models (left). Mice were euthanized at
   day 20 after tumor implantation. Representative histological images
   showing tumor size in Trem2^+/+ (1 and 2) and Trem2^−/− (3 and 4) mice
   (right). Scale bars, 2 mm. (C) Tumor weight. SB28-bearing mice were
   euthanized at day 25. Whole-tumor mass was dissected and weighed (n = 8
   per genotype). (D) Apoptotic cells in tumor tissue by cleaved caspase-3
   (C-caspase 3) immunostaining (n = 3 per genotype). Quantified apoptosis
   shown as mean fluorescence intensity of tumor area (nuclei, DAPI)
   divided by C-caspase 3 intensity (red). Scale bar, 2 mm. (E) Acute
   Trem2 reduction by intracerebroventricular (ICV) injection of ASOs
   increased OS in GBM-bearing Trem2^+/+ mice. SB28, n = 10 per condition;
   NPA C54B, n = 6 per condition. Schematic showing experimental design
   (top). Kaplan-Meier curves demonstrated increased survival in animals
   injected with Trem2 ASO versus control ASO (inactive) (left). Mice were
   euthanized at day 20 after tumor implantation. Representative
   histological images showing tumor size (right). Scale bars, 2 mm. (F)
   Anti-Trem2 treatment increased animal survival in GBM-bearing mice (n =
   5 per condition). Trem2^+/+ mice were injected intracerebrally (IC)
   with SB28 cells and treated intraperitoneally (i.p.) with anti-Trem2 or
   isotype antibody (200 μg per mouse). Experimental design (top).
   Survival curves (bottom). Data represent means ± SEM. *P < 0.05; **P <
   0.01; ***P < 0.001.

   To assess the preclinical effect of Trem2 loss of function on GBM
   growth through an orthogonal approach, we performed
   intracerebroventricular (ICV) injection of Trem2^+/+ mice with
   Trem2-targeted antisense oligonucleotides (ASOs) to acutely knockdown
   Trem2 in the brain, a strategy that is potentially therapeutically
   relevant ([111]25). We first validated that Trem2 ASO treatment
   substantially decreases Trem2 mRNA levels in frontal, parietal,
   occipital, and cerebellar regions of the brain 5 days after injection
   (fig. S2D). The current treatment paradigm of Trem2 ASO by single-dose
   ICV injection was shown to effectively knockdown Trem2 for up to 2
   months in mouse brain ([112]25). We evaluated Trem2 expression in
   tumor-associated myeloid cells 30 days after tumor cell implantation.
   Flow cytometric analysis revealed that both microglial
   (CD45^+CD11b^+Ly6C^low) and blood-lineage myeloid subsets
   (CD45^+CD11b^+Ly6C^high) associated with in vivo SB28 tumors displayed
   reduced Trem2 (fig. S2E). We then asked whether acute, brain-specific
   Trem2 reduction influences GBM tumor growth. Following a single ICV
   dose of Trem2 ASO ([113]Fig. 3E), animals orthotopically injected with
   GBM cells demonstrated significantly increased survival in Trem2
   ASO–treated mice compared to control (inactive) ASO–treated mice using
   both SB28 and NPA C54B cell models ([114]Fig. 3E). Histological
   examination at 20 days after tumor cell injection showed that tumor
   size was diminished in Trem2 ASO–treated mice ([115]Fig. 3E).

   As a third approach to inhibiting Trem2 function, we used a TREM2
   blocking antibody (clone: 178) that was reported to inhibit tumor
   growth in a mouse sarcoma model ([116]15). To better recapitulate the
   patient scenario, TREM2 antibody was administered intraperitoneally 4
   days after orthotopic injection of SB28 tumor cells and then given
   three more times every 4 days ([117]Fig. 3F). Compared to isotype
   antibody, delayed anti-TREM2 antibody treatment also led to improved
   animal survival in SB28-bearing mice ([118]Fig. 3F). Together, these
   convergent results using three different experimental approaches
   indicate that Trem2 loss of function in the TME inhibits tumor growth
   in preclinical models of GBM.

IFNγ-primed Trem2^−/− bone marrow–derived macrophages show increased
phagocytotic function and cytotoxicity against SB28 GBM cells

   To dissect the role of different myeloid subsets in GBM progression, we
   generated bone marrow (BM)–derived macrophages (BMDMs) ([119]26) from
   Trem2 transgenic mice ([120]Fig. 4A and fig. S3, A and B). We first
   evaluated the phagocytic function of BMDMs using fluorescently labeled
   beads ([121]Fig. 4B). Although there was no difference in bead
   phagocytosis between untreated Trem2^+/+ and Trem2^−/− BMDMs, following
   IFNγ treatment, Trem2^−/− BMDMs showed a significant increase in
   phagocytic ability compared with Trem2^+/+ BMDMs ([122]Fig. 4B).
   Macrophage conditioned medium (CM) has been demonstrated to reflect
   macrophage functions ([123]27, [124]28). We then tested the antitumor
   activity of CM derived from primary mouse BMDMs in vitro. Although CM
   from untreated BMDMs did not produce any measurable cytotoxicity in
   SB28 cells (fig. S3C), CM from IFNγ-primed Trem2^−/− BMDMs caused
   notable cytotoxicity to SB28 cells, which was not observed in control
   IFNγ-primed Trem2^+/+ BMDMs ([125]Fig. 4C). We also examined SB28 cell
   proliferation in a Transwell coculture system with no direct tumor cell
   contact with BMDMs. In this system, Trem2^−/− BMDMs substantially
   suppressed tumor cell growth when supplemented with IFNγ compared to
   control Trem2^+/+ BMDMs ([126]Fig. 4D). Consistently, Trem2^−/− BMDMs
   secreted higher levels of nitric oxide (NO) compared to Trem2^+/+ BMDMs
   upon IFNγ treatment ([127]Fig. 4E). We noticed that IFNγ directly
   decreased SB28 proliferation and increased tumor cell NO secretion,
   indicating that the SB28 model is intrinsically susceptible to IFNγ as
   with other cancer cells (fig. S3, D and E) ([128]29, [129]30), but the
   tumoricidal effects of IFNγ were further augmented by activated BMDMs
   in the context of Trem2 deficiency ([130]Fig. 4, C and D). To explore
   the mechanism of how Trem2 deficiency enhances antitumor effects in
   BMDMs, we examined the differentially expressed genes between Trem2^+/+
   and Trem2^−/− BMDMs responding to IFNγ or dead tumor cell stimulation
   using bulk RNA-seq ([131]Fig. 4, F and G). IFNγ robustly induced the
   expression of several proinflammatory genes or M1-type markers in
   Trem2^−/− BMDMs, such as Tnf, Nos2, Cxcl10, Socs3, and Cd80
   ([132]Fig. 4F), consistent with increased tumoricidal function and NO
   secretion in Trem2^−/− BMDMs ([133]Fig. 4, C to E). Subnetwork analyses
   indicated that these differentially expressed genes are highly enriched
   in tumor necrosis factor (TNF) and nuclear factor κB (NFκB) signaling
   pathway ([134]Fig. 4G). In contrast, IFNγ treatment overall suppressed
   the expression of genes associated with the differentiation of MDSCs
   and protumorigenic macrophages ([135]31–[136]33) in Trem2^−/− BMDMs,
   including Cd44, Pkm, Pdk1, Ccr5, and Ccr1 ([137]Fig. 4F). Upon treating
   BMDMs directly with dead GBM cells as antigen, we found much higher
   expression of activated myeloid cell markers including Cd83, H2-T24,
   and Nampt ([138]34) in Trem2^−/− BMDMs ([139]Fig. 4F). Several
   chemokine receptors related with immune tolerance, including Ccr1,
   Ccr3, and Ccr5, were induced at much lower levels in Trem2^−/− BMDMs
   than in Trem2^+/+ BMDMs ([140]Fig. 4F). In addition, we observed a
   dramatically decreased expression of Cd36, Lpl, and Itgam in Trem2^−/−
   BMDMs ([141]Fig. 4F). Pathway analysis showed that these genes regulate
   cholesterol metabolism and phagocytosis ([142]Fig. 4G). In
   tumor-associated immune cells, CD36 has been shown to drive
   lipid-mediated metabolic reprogramming and functions to promote immune
   tolerance and cancer development ([143]35). Lipoprotein lipase (LPL)
   was found to be associated with lipid uptake and an anti-inflammatory
   reparative phenotype in myeloid cells ([144]36). Thus, we speculate
   that impaired lipid reprogramming in Trem2^−/− BMDMs contributes to
   mitigation of the immunosuppressive phenotype. Together, we found Trem2
   deficiency inhibits tolerogenic macrophage differentiation in response
   to both IFNγ and tumor antigen stimulation.

Fig. 4. Activated Trem2^−/− BMDMs show increased phagocytosis and cytotoxic
activity in killing GBM cells and proinflammatory phenotypes.

   [145]Fig. 4.
   [146]Open in a new tab

   (A) Schematic for BMDM induction from both Trem2^+/+ and Trem2^−/−
   littermate mice. BM cells differentiated into macrophages following a
   7-day macrophage colony-stimulating factor (M-CSF) induction. CM
   collected and used for cytotoxic assay as indicated. Gene expression
   was analyzed by bulk RNA-seq. (B) Phagocytosis of fluorescently labeled
   beads measured in BMDMs. BMDMs treated with phosphate-buffered saline
   (PBS) or IFNγ (10 ng/ml) for 30 min. Cytochalasin D (CytoD) was used as
   a negative control. Representative images (left). Scale bar, 100 μm.
   The colocalized ratio of beads in DAPI-positive cells (right; n = 4).
   (C) Cell viability of SB28 assessed by MTS assay. SB28 cells treated
   with CM from IFNγ-primed BMDMs on day 1. Cytotoxic effects examined on
   day 3 (n = 16 pooled from four independent experiments). Abs,
   absorbance. (D) Proliferative curves of SB28 cells. SB28 cells were
   cocultured with BMDMs by the number ratio of 1:2 in a Transwell system
   as shown. Absolute cell numbers of SB28 counted on days 1, 3, and 5 (n
   = 3). (E) Nitrite. BMDMs treated with IFNγ (10 ng/ml) and NO induction
   were assessed (n = 6). (F) Heatmaps of differentially expressed genes
   between Trem2^+/+ and Trem2^−/− BMDMs (n = 2 per condition). BMDMs were
   treated with IFNγ (left) or tumor antigen (right) for 6 hours. (G)
   Enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways in
   cluster 4 upon IFNγ treatment (top, bar plot) and cluster 3 upon dead
   tumor cell stimulation (bottom). NOD, nonobese diabetic. Data represent
   means ± SEM. *P < 0.05; ***P < 0.001.

TREM2 knockdown in human macrophages enhances tumor cell kill in
patient-derived GBM stem cells

   To test the functional role of TREM2, we also took advantage of a human
   model system—patient-derived GBM stem cells (GSCs), which we have
   generated and validated ([147]17, [148]37). Given prior evidence that
   TREM2 might be expressed in primary GBM cells ([149]38), we examined
   TREM2 mRNA levels by real-time quantitative polymerase chain reaction
   (RT-qPCR) in these tumor cells in addition to human neural stem cells,
   human astrocytes, and the human monocyte cell line THP-1, the last as a
   positive control. Except for THP-1 monocytes, none of these cells
   expressed detectable levels of TREM2 (fig. S4A), indicating that TREM2
   marks myeloid cell populations rather than tumor or neural cells,
   consistent with our scRNA-seq data. We further optimized and generated
   THP-1 monocyte-derived macrophage-like cells (fig. S4, B and C), which
   have been used as a valuable model for studying macrophage functions
   ([150]39). We used lentivirus-mediated RNA interference (RNAi) to
   create stable TREM2 knockdown THP-1 macrophages, as well as
   nonsilencing control cells (fig. S5, A to C, and table S1). Assessment
   of cellular morphology and typical macrophage markers revealed no
   significant differences between control and TREM2 knockdown cells (fig.
   S5D), suggesting that TREM2 is not required for THP-1
   monocyte-macrophage differentiation. We next investigated the ability
   of TREM2 to regulate macrophage-mediated cytotoxicity of human GSCs
   ([151]Fig. 5A). Compared to control cells, both short-term (48 hours)
   and long-term (72 hours) CM taken from TREM2 knockdown macrophages in
   culture more robustly decreased GSC viability and increased their
   death, which was observed in two distinct human GSC lines ([152]Fig. 5,
   B to E, and fig. S6, A to D).

Fig. 5. TREM2 knockdown human macrophages demonstrate enhanced cytotoxic
effects on human GSCs.

   [153]Fig. 5.
   [154]Open in a new tab

   (A) TREM2 knockdown THP-1–derived macrophages and their antitumor
   function. Human THP-1 monocytes were stably transduced with
   TREM2-targeted or control short hairpin–mediated RNA (shRNA)
   lentivirus. Macrophage differentiation was induced by 100 nM phorbol
   12-myristate 13-acetate (PMA) for 48 hours. After 24 hours of resting,
   cells were continually cultured for another 48 hours in fresh medium or
   treated with IFNγ (20 ng/ml) for 24 or 48 hours, respectively. CM was
   collected. (B to E) GSCs B152 and B67 treated with CM (48 hours) from
   cultured THP-1 macrophages for 48 hours. Complete medium used as a
   control. (B) and (D) Cell viability shown as luminescence intensity
   (RLU) detected by adenosine 5by ty shown as (ATP) assay. GSCs B152 (n =
   10) and B67 (n = 8). (C) and (E) Cell death shown as positive YOYO-1
   staining (green fluorescence). Representative images (left). Scale
   bars, 100 μm. Death ratio quantified as particle fraction of nuclei
   (Hoechst) divided by YOYO-1 (right). GSCs B152, n = 4; B67, n = 8. scr,
   scrambled; Lum, luminescence. (F and G) Viable cells detected as
   absorbance at 490-nm wavelength by MTS assay. GSCs B152 and B67 treated
   with CM from IFNγ-stimulated THP-1 macrophages for 48 hours. Complete
   medium with IFNγ served the control. (F) GSCs B152 (n = 6 per
   condition). (G) GSCs B67 (n = 6 per condition). Data represent means ±
   SEM. *P < 0.05; **P < 0.01; ***P < 0.001.

   We explored the components in CM from TREM2 knockdown macrophages that
   could induce tumor cell death. We first fractionated the 48-hour CM by
   molecular size and found antitumor activity was present in the 10- to
   100-kDa fraction (fig. S6E). To determine the biochemical nature of the
   cytotoxic activity, we boiled the active fraction to denature proteins,
   which led to the loss of the cytotoxic effect on GSCs (fig. S6E).
   However, the active fraction treated with both ribonuclease (RNase) and
   deoxyribonuclease retained its capacity to induce GSC death (fig. S6E).
   Thus, these data indicate that secreted proteins likely mediate the
   antitumor activity of CM. Given that the proinflammatory cytokine TNFα,
   a 17-kDa protein, is mainly produced by activated macrophages and has
   pleiotropic effects on malignant cells, we examined the production of
   TNFα in CM. TREM2 knockdown macrophages secreted more TNFα than control
   cells in CM (fig. S6F). Furthermore, the cytotoxic effects of CM were
   decreased by addition of adalimumab, an anti-TNFα blocking antibody
   (fig. S6G). We also found that early addition of adalimumab into the
   culture medium of TREM2 knockdown macrophages triggered a more profound
   loss of cytotoxicity of CM compared to control cells (fig. S6H).
   Overall, these results suggest that an autocrine/paracrine TNFα-TNFR
   loop in TREM2 knockdown macrophages leads to secretion of antitumor
   proteins, among which include TNFα.

   Macrophages are generally categorized as M1-like IFNγ-induced
   macrophages and M2-like suppressive macrophages. TREM2 has been shown
   to inhibit mouse macrophage activation ([155]40). We thus investigated
   TREM2’s role in IFNγ-induced adaptive immune responses in THP-1–derived
   human macrophages. IFNγ-treated CM from TREM2 knockdown macrophages
   showed higher cytotoxicity against GSCs under both acute (24 hours) and
   chronic (48 hours) conditions ([156]Fig. 5, F and G). Moreover, upon 24
   hours of IFNγ treatment, the expression of costimulatory molecules
   (CD80 and CD86), proinflammatory cytokines (TNFA and ILB), NOS2, and
   IFN-stimulated genes (ISGs) (CXCL10 and SOCS1) was significantly more
   increased in TREM2 knockdown macrophages, and M2-phenotype gene
   induction such as MRC1 and TGM2 was much lower than in control cells
   (fig. S7, A and B, and table S2). CCL18, which attracts naïve T cells
   and regulatory T cells ([157]41, [158]42), was found to be highly
   up-regulated in TREM2 knockdown macrophages (fig. S7, A and B, and
   table S2), implying homeostatic regulation to avoid excessive
   inflammation. These data suggest that TREM2 acts to hinder
   IFNγ-mediated immunoactivation and TREM2 inhibition promotes M1-like
   phenotype induction upon IFNγ treatment in human macrophages (fig.
   S7C).

   We further examined the role of TREM2 in M2-phenotype polarized
   macrophages upon interleukin-4 (IL-4) and IL-13 treatment (fig. S8A).
   As opposed to the M1-polarized macrophages, we found that CMs from both
   TREM2 knockdown and control macrophages increase GSC viability (fig.
   S8B). Analysis of phenotypic gene expression revealed that mRNA levels
   of M2-phenotype genes were highly induced in control cells, such as
   MRC1, TGM2, CCL18, and IL22, and expression of most M1-pheotype genes,
   including TNFA and IL1B, were not increased in both control and TREM2
   knockdown cells (fig. S8C). Several M2-phenotype genes including IL10,
   TGFB, MRC1, and CCL18 were found to be highly increased in control
   cells compared to TREM2 knockdown cells (fig. S8C). In addition, TREM2
   mRNA levels were strongly induced in control M2-like macrophages (fig.
   S8C). Thus, although TREM2 knockdown restrains M2-phenotype
   polarization in human macrophages, M2-polarized macrophages appear to
   promote tumor growth in a manner independent of TREM2.

Trem2 loss-of-function microglia show increased phagocytic function,
tumoricidal activity, and proinflammatory phenotype

   Microglia are brain-resident macrophages, are distinct from those
   arising from the blood lineage, and have specialized roles in normal
   brain functions and brain diseases ([159]43). We next asked whether
   Trem2-deficient microglial cells might contribute to inhibition of
   tumor growth in vivo ([160]Fig. 3). We used the model of adoptive BM
   transplant to nonconditioned mice, which has advantages over
   myeloablation techniques in that the local environments of organs are
   protected and the hematopoietic niche is preserved. Chimeric BM Trem2
   transgenic mice (CD45.2^+) were generated by adoptive transfer of
   allogeneic BM cells through retro-orbital injection (fig. S9A). Using
   the SB28 GBM model, we found that Trem2^−/− mice with transplanted
   Trem2^+/+ BMs demonstrated a moderate but significant survival
   advantage compared to Trem2^+/+ mice with transplanted Trem2^−/− BMs
   (fig. S9B). Consistently, smaller tumors were observed in SB28-bearing
   Trem2^−/− mice with chimeric BM at day 20 after tumor implantation
   (fig. S9C). We also observed a large quantity of Iba1^+ myeloid
   infiltrates in both Trem2^+/+ and Trem2^−/− mice with chimeric BMs at
   day 25 after tumor injection (fig. S9D), implying a phenotype switch
   rather than a change in myeloid cell number. These results suggest that
   Trem2-deficient resident microglia also play a substantial role in
   tumor suppression in vivo. To better understand the antitumor role of
   Trem2-deficient microglia, we used the BV2 cell line, a commonly used
   model of microglial function ([161]44). Trem2 knockdown in BV2 cells
   was confirmed using both lentiviral transduction and Trem2 ASO
   treatment ([162]Fig. 6, A and C). We measured the phagocytic activity
   in both RNAi-transduced BV2 cells and ASO-treated BV2 cells using an in
   vitro bead assay. Under both conditions, Trem2 knockdown BV2 cells
   showed a significant increase in their ability to engulf fluorescently
   labeled beads compared with control cells ([163]Fig. 6, B and D). We
   then isolated and cultured primary microglia ([164]45) from Trem2
   transgenic mice ([165]Fig. 6E). Similar to BV2 cells, Trem2^−/− primary
   microglia showed increased phagocytotic capacity of fluorescently
   labeled beads ([166]Fig. 6E).

Fig. 6. Trem2-deficient microglial cells demonstrate increased phagocytotic
activity.

   [167]Fig. 6.
   [168]Open in a new tab

   (A to D) Phagocytosis of fluorescently labeled beads was measured in
   BV2 cells. (A) Trem2 expression. Trem2 knockdown in BV2 cells by
   lentivirus transduction: Trem2-targeted shRNA (Trem2i.1 and Trem2i.2)
   and nonsilencing shRNA (scr). Fold change of Trem2 mRNA (left; n = 6).
   Representative Western blot (right, n = 2). (B) Phagocytotic activity.
   Cytochalasin D, an inhibitor of actin polymerization, was used as a
   negative control. Representative images (left). Scale bar, 100 μm. The
   colocalized ratio of beads in DAPI-positive cells (right; n = 6). (C)
   Trem2 expression. BV2 cells treated with inactive or Trem2-targeted ASO
   for 48 and 72 hours, respectively. Fold change of Trem2 mRNA (left; n =
   4). Representative Western blot (right; n = 2). (D) Phagocytotic
   activity. Cytochalasin D served the control. Representative images
   (left). Scale bar, 100 μm. Quantified colocalization of beads with
   DAPI-positive cells (right; n = 8). (E) Phagocytosis of fluorescently
   labeled beads measured in primary microglia. Primary microglia isolated
   from trem2 transgenic mice. Representative images (left). Scale bar,
   100 μm. The colocalized ratio of beads in DAPI-positive cells (right; n
   = 41 fields per condition). Data represent means ± SEM. *P < 0.05; **P
   < 0.01; ***P < 0.001.

   We next examined BV2-mediated anti-GBM function using CM from naïve or
   IFNγ-polarized BV2 cells ([169]Fig. 7A). Compared to control cells,
   both short-term (48 hours) and long-term (96 hours) CM from Trem2
   knockdown BV2 cells demonstrated greater cytotoxicity on SB28 GBM cells
   ([170]Fig. 7B). CM from IFNγ-treated control BV2 cells led to higher
   tumor cell kill in SB28 cells compared to untreated CM ([171]Fig. 7B).
   In addition, CM from IFNγ-treated Trem2 knockdown BV2 cells showed
   greater antitumor effects compared to that of IFNγ-treated control
   cells ([172]Fig. 7B). We also monitored SB28 cell proliferation and
   viability through indirect contact coculture experiments with BV2
   microglia ([173]Fig. 7C). Trem2 knockdown BV2 cells significantly
   inhibited the proliferation and viability of SB28 tumor cells with the
   addition of IFNγ ([174]Fig. 7D). We further isolated primary microglia
   from Trem2^+/+ mice and treated these cells with ASOs. After Trem2 ASO
   treatment, mRNA and protein levels of Trem2 in primary microglia were
   markedly decreased (fig. S10A). As in our BV2 experiments, we collected
   CM from ASO-treated primary microglia. Both short-term (48 hours) and
   long-term (72 hours) CM from Trem2 ASO–treated primary microglia
   demonstrated greater cytotoxicity on SB28 cells than control
   ASO–treated cells ([175]Fig. 7E). We then analyzed the antitumor
   effects of CM from IFNγ-treated primary microglia on tumor cells. Both
   Trem2 ASO– and control ASO–treated primary microglia could be activated
   by IFNγ, showing increased phosphorylation in signal transducers and
   activators of transcription 1 (STAT1) (fig. S10B). CM from IFNγ-treated
   microglia with reduced Trem2 was more effective in suppressing tumor
   cell growth compared to that of control treated cells ([176]Fig. 7F).

Fig. 7. Trem2-deficient microglia exhibit increased tumoricidal activity and
proinflammatory phenotype.

   [177]Fig. 7.
   [178]Open in a new tab

   (A to D) Trem2 knockdown BV2 microglia were generated using
   Trem2-targeted shRNA lentivirus transduction (Trem2i.1 and Trem2i.2).
   Nonsilencing shRNA transduced cells were used as control (scr). (A)
   Schematic for the experimental strategy. CM was collected from
   untreated or IFNγ-treated BV2 cells. The cytotoxicity of CM on SB28
   cells was assessed. (B) Cell viability detected as luminescence
   intensity. Top, untreated; bottom, IFNγ-treated; left, 48-hour CM;
   right, 96-hour CM (n = 12 pooled from three independent experiments).
   (C) and (D) Indirect antitumor effects of BV2 microglia assessed. (C)
   Experimental design for Transwell BV2-SB28 coculture. (D) Proliferative
   curve and cell viability. SB28 proliferation detected by absolute cell
   count (left; n = 3). Viable cells examined by ATP assay (right; n = 4).
   (E to H) Primary microglia treated with inactive (control) or
   Trem2-targeted ASOs for 72 hours. (E) The cytotoxicity of CM on SB28
   cells assessed by MTS assay. CM collected from primary microglia
   following ASO treatment. Left, 48-hour CM (n = 8); right, 72-hour CM (n
   = 8). (F) Cytotoxic effects of IFNγ-treated CM on SB28 cells. After ASO
   treatment, primary microglia were stimulated with IFNγ (10 ng/ml) for
   24 or 48 hours, and CM was collected (n = 8). (G) Heatmaps of
   differentially expressed genes between inactive and Trem2 ASO–treated
   primary microglia (n = 2 per condition). Primary microglia were treated
   with IFNγ (left) or dead SB28 cells (right) for 6 hours. (H) KEGG
   pathway enrichment analysis of cluster 1. Networks: IFNγ treatment
   (top) and tumor antigen treatment (bottom). Data represent means ± SEM.
   *P < 0.05; **P < 0.01; ***P < 0.001.

   To elucidate the mechanism of these findings, we first examined the
   resting microglial phenotype in BV2 (table S3). Upon Trem2 knockdown,
   the expression of a panel of microglial cell surface maker genes were
   not altered, including Cd38, Cd68, Cd169, Cd204, Mrc1, Tmem119, P2ry12,
   and Cx3cr1 (fig. S10C). Following IFNγ treatment, both control and
   Trem2 knockdown BV2 cells converted into amoeboid-like cells and the
   proliferative rates were similarly decreased (fig. S10, D and E).
   However, upon IFNγ stimulation, Trem2 knockdown augmented M1-phenotype
   polarization in BV2 cells: gene expression involved in antigen
   presentation (Cd80 and Cd86), immune cell recruitment (Ccl2 and Cxcl9),
   and inflammation (Il6 and Nos2) was more increased; expression of
   Cx3cr1, Mrc1, and Arg1 was further decreased compared to control cells
   (fig. S10F). Moreover, TNFα and NO secretion in Trem2 knockdown BV2
   cells was higher than in control cells following IFNγ treatment (fig.
   S10, G and H). Together, Trem2 loss of function up-regulates
   IFNγ-induced proinflammatory potential in BV2 microglia.

   To understand the antitumor phenotype of Trem2 loss-of-function
   microglia in a more unbiased fashion, we analyzed the transcriptomic
   profiles of ASO-treated primary microglia by RNA-seq ([179]Fig. 7, G
   and H). Upon IFNγ treatment, Trem2 ASO–treated cells showed much lower
   induction of Cd274 and Batf2 than inactive ASO-treated cells
   ([180]Fig. 7G), suggesting stronger activation of immunosuppressive
   pathways ([181]46) in control primary microglia ([182]Fig. 7H). The
   expression of Gsn, Clec7a, and Plcg2 was markedly inhibited in Trem2
   ASO–treated primary microglia ([183]Fig. 7G). Gelsolin (GSN) plays an
   anti-inflammatory role by binding to inflammatory mediators, and
   phospholipase C gamma 2 (PLCG2) is required for TREM2-initiated
   signaling in microglia ([184]47). Dectin-1 (Clec7a) drives
   immunosuppressive macrophage differentiation in the TME ([185]48).
   Moreover, compared to control cells, gene expression of Ctsz, Npc2, and
   Atf4 was highly induced in Trem2 ASO–treated primary microglia
   ([186]Fig. 7G). Cathepsin Z (CTSZ) and NPC2 ([187]49) play roles in
   regulating endosomal/lysosomal function, such as cholesterol
   trafficking and antigen presentation, and are highly expressed in
   M1-type macrophages. Activating transcription factor 4 (ATF4), a
   transcription factor, participates in proinflammatory cytokine and
   chemokine production, such as IL-6 and monocyte chemoattractant
   protein-1 (MCP-1) ([188]50). Together, Trem2 ASO–treated primary
   microglia demonstrated a much more immunoactive phenotype upon IFNγ
   stimulation. We further treated primary microglia with dead tumor cells
   to explore their responses to tumor antigens ([189]Fig. 7G). We found
   that Trem2 ASO–treated microglia exhibited relatively low expression of
   Plau, Itgax, and Itgam ([190]Fig. 7G). Kyoto Encyclopedia of Genes and
   Genomes (KEGG) pathway analysis indicated that these genes are involved
   in the NFκB signaling pathway, phagocytosis of apoptotic cells or
   immune complexes, and cancer progression ([191]Fig. 7H). Upon tumor
   cell treatment, we found a highly induced cluster of genes in Trem2
   ASO–treated microglia, including H2-D1, Atf4, Sqstm1 ([192]51), and
   Mcl1 ([193]52) ([194]Fig. 7H), which are associated with increased
   capacity for antigen presentation, cytokine induction, and
   proinflammatory phenotype polarization. Conversely, compared to control
   microglia, Trem2 ASO–treated microglia showed much lower expression of
   Cd44, Cd33, Ccl9, Mmp9, Ifnar1, and Tlr2 ([195]Fig. 7H), most of which
   are related with M2-type macrophage polarization and immunosuppressive
   myeloid subset induction. Together, these data demonstrate Trem2
   inhibition promotes immunoactive microglial differentiation.

Trem2 loss-of-function microglial cells decrease tumor growth in coculture
and tumor initiation in vivo

   We next evaluated the direct tumoricidal activity of Trem2
   loss-of-function myeloid cells both in vitro and in vivo ([196]Fig. 8).
   We first evaluated the antitumor activity of Trem2 knockdown BV2
   microglial cells, using a coculture system ([197]Fig. 8A). We observed
   that coculture of control BV2 cells with SB28 cells led to substantial
   expansion of tumor cells and little, if any, growth of BV2 microglia by
   day 5 ([198]Fig. 8B). However, coculture of Trem2 knockdown BV2 cells
   with SB28 cells resulted in notable inhibition of tumor cell growth and
   expansion of microglial cells ([199]Fig. 8, B and C). To understand how
   Trem2 knockdown increases tumor cell kill and microglial expansion, we
   first examined TNFα production in the coculture system and found higher
   TNFα secretion in cocultures using Trem2 knockdown BV2 cells (fig.
   S11A). We then analyzed the expression of B-cell lymphoma 2 (Bcl-2) and
   Bcl-2 associated X (Bax) in BV2 cells, which reflect cell intrinsic
   survival signals ([200]53, [201]54). We found Trem2 knockdown BV2 cells
   exhibit a much higher ratio of Bcl-2 to Bax than control cells,
   suggesting that Trem2 inhibition enhances the survival capacity of BV2
   microglial cells (fig. S11B). We also cocultured primary microglia with
   SB28 cells in vitro and similarly found that the growth rate of tumor
   cells cocultured with Trem2^−/− microglia was lower than those
   cocultured with Trem2^+/+ microglia (fig. S11C), in line with the
   results with BV2 cells.

Fig. 8. Trem2 loss-of-function myeloid cells show antitumor effects both in
vitro and in vivo.

   [202]Fig. 8.
   [203]Open in a new tab

   (A to C) Trem2 knockdown in BV2 microglia by Trem2-targeted shRNA
   lentivirus transduction (Trem2i.1 and Trem2i.2). Nonsilencing shRNA
   used as control (scr). BV2 and SB28 cells were stained with green and
   deep-red tracking dye, respectively, and cocultured in vitro. (A)
   Timeline schematic for treatment strategy. (B) Representative images.
   Green, BV2; purple, SB28; blue, nucleus (DAPI). Scale bar, 100 μm. (C)
   Violin plot indicating the ratio of mean fluorescent intensity
   (green/purple) at day 5 (quantified for 30 fields per condition). (D)
   GBM model of intracranial (IC) coimplantation of BV2 microglia and SB28
   cells. Neurological deficit-free survival curves (right; n = 5 per
   group). WT, wild-type. (E) GBM model of BMDMs and SB28 coinjection.
   Neurological deficit-free survival curves (right; n = 4 per group).
   Data represent means ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001.

   We then investigated the signaling pathways that Trem2 may modulate in
   microglial cells using BV2 cells. Upon IFNγ treatment, we found STAT1
   activation was decreased in Trem2 knockdown BV2 microglia (fig. S12, A
   and B). Moreover, altered mitogen-activated protein kinases (MAPKs) and
   NFκB pathway activation was observed in Trem2 knockdown BV2 microglia
   upon lipopolysaccharide (LPS) treatment (fig. S12C). Both extracellular
   signal–regulated kinase (ERK) and c-Jun N-terminal kinase (JNK)
   activation were decreased, while MAPK p38 activity was increased in
   Trem2 knockdown BV2 microglia (fig. S12, C and D). Phosphorylated
   inhibitor of NFκBα (IκBα) demonstrated an increase at baseline in Trem2
   knockdown BV2 cells although stimulated activity was unchanged,
   suggesting that Trem2 inhibits ambient but not LPS-dependent NFκB
   activity (fig. S12, C and D). We further examined MAPKs activation in
   BV2 cells using whole tumor cell antigen. We found dead SB28 cells did
   not activate microglial ERK and JNK but increased activation of NFκB
   and MAPK p38 in Trem2 knockdown BV2 cells compared to control cells
   (fig. S12, E and F). Therefore, Trem2 remodels microglial function by
   participating in multiple signaling pathways.

   Given the antitumor effects of Trem2 loss-of-function macrophages and
   microglia using CM and the direct coculture system, we predicted that
   Trem2 loss-of-function myeloid cells could potentially kill tumor cells
   and inhibit tumor growth in vivo if present at the early stage of tumor
   initiation. We therefore designed an in vivo coimplantation model in
   which we inject SB28 GBM cells together with myeloid cells. We found
   that orthotopic coinjection of SB28 cells with control BV2 microglial
   cells did not alter animal survival compared to injection of tumor
   cells alone ([204]Fig. 8D). However, coinjection of SB28 cells with
   Trem2 knockdown BV2 microglia significantly increased animal survival
   compared to injection of tumor cells with control cells ([205]Fig. 8D).
   We found similar results with coinjection of SB28 cells with Trem2^−/−
   BMDMs ([206]Fig. 8E). Together, these results suggest that Trem2
   loss-of-function myeloid cells have direct antitumor capacity in vivo.

Trem2 deficiency reprograms tumor-associated myeloid populations toward a
proinflammatory phenotype

   Our data indicate that Trem2-deficient myeloid cells exhibit enhanced
   antitumor activity and proinflammatory phenotypes in vitro and curb
   tumor growth in vivo using coinjection models. We next analyzed the
   phenotypes of tumor-associated myeloid cells in vivo in the context of
   Trem2 loss of function. We first examined the myeloid infiltrates of
   GBM tumors in Trem2 transgenic mice. We found large numbers of Iba1^+
   (pan-myeloid) and transmembrane protein 119 (TMEM119)^+ (microglial)
   myeloid infiltrates in SB28 orthotopic tumors in both Trem2^−/− and
   Trem2^+/+ mice ([207]Fig. 9A). No significant differences were observed
   in frequencies of both local and migrated myeloid subsets among total
   myeloid infiltrates between Trem2^+/+ and Trem2^−/− mice
   ([208]Fig. 9B). We then looked at inflammation-related gene expression
   (table S3) in intratumoral CD45^+CD11^+ cells that were sorted using
   flow cytometry ([209]Fig. 9C). Expression of the typical
   proinflammatory genes Nos2 and Cxcl9 was higher in Trem2^−/− myeloid
   infiltrates than in Trem2^+/+ myeloid infiltrates in tumors
   ([210]Fig. 9C). In contrast, expression of the anti-inflammatory gene
   Arg1 was higher in Trem2^+/+ myeloid infiltrates than in Trem2^−/−
   infiltrates ([211]Fig. 9C).

Fig. 9. Trem2 deficiency reprograms tumor-associated myeloid cells toward a
proinflammatory phenotype.

   [212]Fig. 9.
   [213]Open in a new tab

   (A) Representative imaging showing that Iba1^+ and Tmem119^+ myeloid
   cell populations in SB28 tumor from Trem2^+/+ and Trem2^−/− mice. Scale
   bars, 100 μm. (B) Representative pseudo-color/smooth plots showing
   intratumoral CD45^+CD11b^+ myeloid cells (top). Frequencies of
   Ly6C^high (macrophage/monocyte lineage) and Ly6C^low cells (resident
   microglia) among CD45^+CD11b^+ cell populations (bottom; n =
   8/genotype). (C) Relative gene expression in tumor-associated
   CD45^+CD11b^+ myeloid cells from Trem2^+/+ and Trem2^−/− mice (n = 4
   per genotype). (D to G) Trem2^+/+ mice were intracerebroventricularly
   injected with Trem2-targeted ASO or inactive control. Then, ASO-treated
   mice were intracerebrally implanted with SB28 cells. Live CD45^+ cells
   isolated from brain tumors (day 30) and analyzed by scRNA-seq. (D) and
   (F) Volcano plots showing differentially expressed genes in macrophage
   subset (D) and microglia population (F) from Trem2 ASO–treated tumor
   versus from inactive ASO–treated tumor. (E) KEGG analysis of enriched
   up-regulated and down-regulated pathways in tumor-associated macrophage
   subset. Red, up-regulated; blue, down-regulated. (G) STRING protein
   interaction analysis of down-regulated genes (Gpnmb and Lpl) and
   up-regulated genes (Tnfsf4 and Dapk2) in tumor-associated microglia
   from Trem2 ASO–treated tumor. Networks generated on STRING v11.5
   ([214]https://string-db.org/). Data represent means ± SEM. *P < 0.05;
   **P < 0.01. NS, no significance; FC, fold change.

   We then focused on the phenotypes of tumor-associated myeloid cells
   from Trem2 ASO-treated mice. Using scRNA-seq (fig. S13), we compared
   the differential gene expression of myeloid infiltrates between Trem2
   ASO– and inactive ASO–treated mice ([215]Fig. 9, D to G). For
   tumor-associated macrophages, we focused on the lipid
   metabolism–associated macrophage subset since we observed the most
   robust gene expression changes in Trem2 ASO–treated mice. We observed
   significantly up-regulated proinflammatory markers (Ccl2, Ccl12, Cxcl3,
   and Cxcl10), macrophage activation markers (Il18 and Cyp2s1) ([216]55),
   and IFN-induced protein (Ifit3) ([217]Fig. 9D) ([218]56). Ccl24, which
   mediates proinflammatory neutrophil and macrophage chemotaxis, was also
   increased in expression ([219]Fig. 9D). Oppositely, genes that
   contribute to the resolution of inflammation and tissue repair
   ([220]57) were significantly decreased in expression, such as Mgp,
   Hsd11b1, Gstm2, and Plet1 ([221]Fig. 9D). The expression of Pla2g2d, a
   M2-type macrophage marker ([222]58), was also decreased ([223]Fig. 9D).
   Treml4, a Trem family member that regulates inflammatory responses,
   decreased in expression as well ([224]Fig. 9D). Furthermore, in Trem2
   ASO–treated mice, this tumor-associated macrophage subset demonstrated
   significant up-regulation of the TNF and IL-17 signaling pathways
   ([225]Fig. 9E), which are involved in tumor cell killing and
   proinflammatory responses. We also observed a decrease in ferroptosis
   pathway genes ([226]Fig. 9E), which leads to lipid peroxidation and
   eventual cell death. In tumor-associated microglia of Trem2 ASO–treated
   mice, we observed a different set of up-regulated genes that contribute
   to a proinflammatory phenotype ([227]Fig. 9, F and G): Tnfsf4, Il12rb2,
   and Chrm3 are involved in proinflammatory response and cytokine
   induction; Rsad2 is one of the most highly induced IFN effector genes;
   Ctsw and Dapk2 ([228]59) modulate lysosomal function and autophagy;
   Gzma and Pdpn play roles in phagocytosis and production of
   proinflammatory cytokines ([229]60, [230]61). As in tumor-associated
   macrophages, we observed decreased expression of Mgp and Plet1 in the
   tumor-associated microglial subset of Trem2 ASO–treated mice
   ([231]Fig. 9F). Gpnmb and Grem1, which inhibit macrophage M1
   polarization ([232]62, [233]63), as well as Lpl and Scara3 ([234]64),
   which correlate with dysregulated lipid metabolism and
   anti-inflammatory responses, were down-regulated in microglia from
   Trem2 ASO–treated mice ([235]Fig. 9, F and G). Dcbld1, associated with
   poor prognosis in patients with cancer ([236]65), was also found to be
   down-regulated in tumor-associated microglia from Trem2 ASO–treated
   mice ([237]Fig. 9F). Protein-protein network analyses (String) suggest
   that transmembrane glycoprotein NMB (Gpnmb) and Trem2 are functionally
   linked ([238]Fig. 9G), tumor necrosis factor superfamily, member 4
   (Tnfsf4) delivers costimulatory signals between microglia and T cells
   ([239]Fig. 9G), and death-associated protein kinase 2 (Dapk2) regulates
   MAPK activity in microglia ([240]Fig. 9G), implying that Trem2
   regulates a wide range of intracellular and cell-cell signaling
   interactions. Together, tumor-associated myeloid cells in the setting
   of Trem2 loss of function demonstrate enhanced proinflammatory
   functions.

Trem2 deficiency increases PD-1^+CD8^+ T cells in the TME, and combined
anti–PD-1 treatment improves animal survival

   Trem2 loss of function reprograms the phenotypes and functions of
   myeloid infiltrates in the GBM TME. Given that cytotoxic T cells (CD8^+
   T cells) are thought to represent a major immune subset that can kill
   tumor cells and inhibit tumor growth, we asked whether Trem2 deficiency
   affects T cell functions in GBM. T cells constitute a very small
   cellular fraction of the GBM TME, and suppressive myeloid cells use
   multiple mechanisms to inhibit T cell responses. We examined
   tumor-associated lymphoid infiltrates using ASO-treated mice bearing
   SB28 tumors. Trem2 ASO–treated mice demonstrated tumor infiltrates with
   higher frequencies of total T cells and CD8^+ T cells than control
   ASO–treated mice ([241]Fig. 10A), implying a more robust T cell
   response in Trem2 ASO–treated mice. Moreover, Trem2 ASO–treated mice
   exhibited more T cells that express programmed cell death protein 1
   (PD-1) in tumor infiltrates, especially CD8^+ T cells ([242]Fig. 10A),
   suggesting increased T cell activation in the TME and potentially
   enhanced responsiveness to anti–PD-1 checkpoint blockade. We also
   examined PD-1 expression in CD8^+ T cells of tumor infiltrates from
   Trem2 transgenic mice bearing SB28 tumors. Both the frequency of PD-1^+
   T cells among CD8^+ T cells ([243]Fig. 10B) and average PD-1 protein
   level in all CD8^+ T cells ([244]Fig. 10C) were higher in tumors from
   Trem2^−/− mice compared to Trem2^+/+ mice. Collectively, these results
   suggest that Trem2 deficiency remodels myeloid infiltrates in a manner
   that enhances T cell activation, likely contributing to restricted
   tumor growth in vivo.

Fig. 10. Trem2 deficiency increases PD-1^+CD8^+ T cells in the TME, and
combined anti–PD-1 therapy augments overall animal survival.

   [245]Fig. 10.
   [246]Open in a new tab

   (A) Representative pseudo-color/smooth plots showing intratumoral T
   cell subsets and PD-1 expression (top). Live CD45^+ cells isolated from
   SB28 tumors in ASO-treated mice and assessed by flow cytometry. Summary
   graphs showing frequencies of total T cells among CD45^+ cells,
   CD3^+CD8^+ and CD3^+CD4^+ subsets among T cells, and PD-1^+ cell among
   different T cell populations, and PD-1 protein levels [mean
   fluorescence intensity (MFI)] (bottom; n = 8 per condition). (B to C)
   Tumor-associated CD45^+ cells isolated from SB28-bearing Trem2^+/+ and
   Trem2^−/− mice. PD-1 expression in CD3^+CD8^+ T cells was assessed by
   flow cytometry. (B) Frequency of PD-1^+ cells in CD3^+CD8^+ T cells.
   Representative flow plots (left) and summary graph (right; n = 8 per
   genotype). (C) PD-1 protein shown as MFI. Representative histograms
   (left) and summary graph (right; n = 8 per genotype). TIL,
   tumor-infiltrating lymphocyte. (D and E) Combined therapy in GBM. (D)
   Schematic (top). Trem2^−/− mice were treated with anti–PD-1 antibody or
   isotype control following tumor implantation (n = 5 per condition).
   Symptom-free survival curves for SB28-bearing mice (bottom). (E)
   Schematic showing ASOs and anti–PD-1 treatment strategy (top).
   Neurological deficit-free survival curves for SB28 GBM model (bottom).
   (F) Schematic to describe Trem2 inhibition reshapes TME by modulating
   myeloid cell phenotypes and cytotoxic T cell function, which suppresses
   tumor progression. Data represent means ± SEM. *P < 0.05; **P < 0.01;
   ***P < 0.001. IC, intracranial; ICV, intracerebroventricular.

   Given these alterations in T cell number and phenotype, we then tested
   whether we could improve the efficacy of Trem2 deficiency in our tumor
   model with the addition of an immune checkpoint inhibitor, anti–PD-1
   antibody ([247]Fig. 10, D and E). We first tested the efficacy of
   anti–PD-1 immunotherapy in combination with Trem2 transgenic mice in
   our tumor model ([248]Fig. 10D). We found that compared to isotype
   treatment, anti–PD-1 administration significantly prolonged overall
   animal survival in SB28-bearing Trem2^−/− mice despite initiating
   anti–PD-1 8 days following tumor injection ([249]Fig. 10D). We also
   tested the ability of PD-1 in combination with Trem2 ASO treatment. It
   is known that the SB28 GBM model is completely resistant to anti–PD-1
   treatment ([250]19, [251]66). Consistently, anti–PD-1 therapy (with
   control ASO) had no effect on animal survival in SB28-bearing mice
   ([252]Fig. 10E). Compared to Trem2 ASO treatment alone, Trem2 ASO
   combined with anti–PD-1 treatment increased OS in SB28-bearing mice
   ([253]Fig. 10E). Therefore, we conclude that TREM2 deficiency in the
   TME not only reprograms myeloid cell function but also promotes T cell
   infiltration and activation, which intensifies anti–PD-1 therapy
   ([254]Fig. 10F). Together, Trem2 inhibition in myeloid infiltrates,
   including microglia and blood-lineage macrophages, reshapes the
   microenvironment toward a more immunologically “hot” status that
   suppresses GBM growth and leads to improved animal survival.

DISCUSSION

   In this study, we identified that higher expression of TREM2 in
   IDH1/2–wild-type GBMs is associated with worse survival in patients
   with GBM and TREM2 expression is enriched in myeloid cell populations
   in GBM TME, including microglia and macrophage subsets. To explore the
   role of Trem2 in IDH–wild-type GBM, we used two syngeneic orthotopic
   mouse GBM models, SB28 ([255]22) and NPA C54B ([256]24). Using
   scRNA-seq and immunostaining, we found abundant Trem2^+ myeloid
   infiltrates in the TME during GBM progression in vivo. We then used
   Trem2 transgenic mice and found Trem2 loss of function in vivo leads to
   increased tumor cell apoptosis, slower tumor expansion, and improved
   animal survival in GBM-bearing mice. Previously, acute Trem2 reduction
   by anti-TREM2 antibody treatment was found to be more effective than
   TREM2 knockout in controlling tumor growth in a mouse sarcoma model
   ([257]15). Considering the brain context of GBM, we further used ICV
   injection of Trem2-lowering ASOs to acutely knockdown Trem2 expression
   in mouse brain. We found that acute, brain-specific Trem2 reduction
   robustly reduced tumor size and increased animal survival in two GBM
   models. We corroborated this with anti-TREM2 antibody treatment, which
   also improved animal survival in GBM-bearing mice.

   To understand the mechanism of Trem2 inhibition in regulating GBM
   progression, we focused on the Trem2-expressing populations in GBM TME,
   including blood-lineage macrophages and resident microglia. First, we
   found in vitro that IFNγ-induced Trem2^−/− BMDMs demonstrated increased
   phagocytic and tumor-suppressing capacity than Trem2^+/+ BMDMs. Using
   bulk RNA-seq, we found that IFNγ-treated Trem2^−/− BMDMs express a much
   higher level of M1-related, proinflammatory genes including Tnf, Nos2,
   Ccl5, and Cxcl10, which are highly enriched in TNF and NFκB signaling
   pathway, consistent with the increased tumoricidal activity of
   Trem2^−/− BMDMs. We also stimulated BMDMs with tumor cell antigen to
   detect their antitumor responses. We found that upon both IFNγ and
   tumor antigen treatment, Trem2^−/− BMDMs were more immunoactive by
   expressing higher levels of activated myeloid cell markers such as
   Cd80, Cd83, H2-T24, and Nampt and lower levels of chemokine receptors
   related with immune tolerance, such as Ccr1 and Ccr5. Moreover,
   Trem2^−/− BMDMs showed much higher expression of genes related with
   endosomal/lysosomal functions, such as Npc1, Ctsd, Ctsz, Atp6v1a,
   Lamp1, and Lamtor3, contributing to intracellular antigen processing
   and presentation. Lipid uptake and metabolism, associated with
   anti-inflammatory phenotypes in myeloid cells, were significantly
   down-regulated in Trem2^−/− BMDMs. In addition, we found that TREM2
   knockdown human macrophages exhibited more cytotoxic effects on human
   GSCs than control macrophages. In human macrophages, TREM2 was also
   found to restrict IFNγ-induced immunoactivation and proinflammatory
   phenotypes. We identified TNFα as one of the secreted antitumor factors
   in TREM2 knockdown human macrophages and provide evidence for
   additional secreted antitumor proteins, which remain to be found.
   Moreover, TREM2 knockdown restrained human macrophage M2 polarization.
   Together, TREM2 inhibition can remodel macrophages toward an
   immunoactive functional state in vitro.

   Our results using BM chimeric Trem2 transgenic mice in the SB28 GBM
   model revealed that Trem2 loss of function in brain microglia also
   played an important role in tumor suppression in vivo. We further
   examined how Trem2 regulates microglial function by using the BV2 cell
   line ([258]44) and primary microglia ([259]45). We found that Trem2
   loss of function in both BV2 cells and primary microglia boosted
   phagocytic capacity. Both Trem2 knockdown BV2 cells and Trem2
   ASO–treated primary microglia showed augmented tumor–suppressing
   function. Gene expression profiling of primary microglia upon IFNγ or
   tumor antigen stimulation revealed that Trem2 inhibition significantly
   inhibited the expression of genes that down-regulate inflammation or
   promote immunosuppression, such as Gsn, Clec7a, Plcg2, Cd44, Cd33,
   Plau, and Ccl9, and highly induced the expression of genes involved in
   antigen presentation, cytokine induction, or proinflammatory
   phenotypes, including H2-D1, Ctsz, Npc2, Atf4, Sqstm1, and Mcl1. Thus,
   Trem2 inhibition promotes immunoactive microglial differentiation.

   To further validate that Trem2 inhibition enhances antitumor activities
   in both macrophages and microglia, we performed direct coculture of
   myeloid and GBM cells in vitro and a coimplantation model using primary
   BMDMs or BV2 cells with GBM cells in vivo. We found, both in vitro and
   in vivo, that Trem2 loss-of-function myeloid cells demonstrate
   tumor-suppressing function. These data indicate that Trem2 loss of
   function in the TME inhibits tumor growth and further suggests that
   myeloid phenotypes rather than the absolute number of myeloid cell
   populations in the TME affect tumor growth.

   To expand our knowledge of how Trem2 regulates classical signaling of
   myeloid cells, we used the BV2 model. We found that Trem2 was involved
   in multiple signaling pathways in BV2 cells. Trem2 knockdown decreased
   STAT1 overactivation upon IFNγ treatment, which may prevent sustained
   expression of apoptosis-related ISGs and lead to increased microglial
   survival ([260]67, [261]68). Moreover, Trem2 knockdown remodeled
   LPS-induced activation of MAPKs signaling, including decreased ERK and
   JNK activation and increased MAPK p38 activation. Essential roles for
   the MAPKs in macrophage function have been established: ERK signaling
   contributes to the M2-type macrophage development and anti-inflammatory
   cytokine induction ([262]69, [263]70); Activation of JNK family members
   is important for M1 polarization and macrophage survival ([264]71);
   MAPK p38 pathway is activated in M1 macrophages and supports cell
   survival as well as the production of proinflammatory mediators, such
   as TNFα ([265]70, [266]72). In this regard, we found that activation of
   MAPK p38 correlates with enhanced proinflammatory response in Trem2
   knockdown BV2 cells. Moreover, Trem2 knockdown elevated NFκB activation
   at baseline, which may, in part, explain why CM from Trem2 knockdown
   cells more potently inhibit the proliferation of GBM cells. We examined
   MAPK activation in BV2 cells that have been stimulated with dead tumor
   cells. In this context, we did not observe activation of ERK and JNK in
   both control and Trem2 knockdown microglia. However, Trem2 knockdown
   BV2 cells exhibited significantly increased NFκB and MAPK p38
   activation, both of which induce genes that inhibit cell apoptosis and
   enhance inflammation ([267]73).

   Considering the complexity of GBM TME, we further investigated the
   phenotypes of tumor-associated myeloid cells in vivo by flow cytometry
   and scRNA-seq. We found tumor-associated Trem2^−/− myeloid cells more
   highly express the M1-phenotype genes and more lowly express the
   anti-inflammatory gene Arg1. Using acute Trem2 knockdown GBM model, we
   found that proinflammatory molecules including Ccl2, Ccl12, Cxcl3, and
   Cxcl10 and the TNF signaling pathway are significantly up-regulated in
   Trem2-inhibited in the lipid metabolism-associated macrophage subset,
   in line with what we have observed in vitro. Tumor-associated microglia
   with Trem2 reduction demonstrated increased lysosomal function and
   autophagy (Ctsw and Dapk2), consistent with our in vitro results using
   cultured primary microglia (Ctsz and Npc2). Moreover, the expression of
   genes (Tnfsf4, Il12rb2, Pdpn, and Chrm3) that promote proinflammatory
   response and cytokine induction were increased tumor-associated
   microglia in the context of Trem2 deficiency. On the contrary, we
   observed that the gene expression profiles that contribute to
   anti-inflammatory responses and tissue homeostasis were down-regulated
   in both Trem2-inhibited tumor-associated macrophages and microglia. For
   example, Mgp and Plet1 expression was decreased in both the macrophage
   subset and microglia. Overall, tumor-associated myeloid cells with
   Trem2 loss of function demonstrated enhanced proinflammatory functions,
   corroborating our in vitro findings.

   Furthermore, we found molecules that play roles in lipid uptake and
   metabolism were significantly inhibited in expression in Trem2
   loss-of-function myeloid cells (macrophage and microglia) both in vitro
   and in vivo, such as Cd36, LPL, phospholipase A2 group IID (PLA2G2D),
   PLCG2, Treml4, and scavenger receptor class A member 3 (SCARA3). It has
   been reported that TREM2 is a lipid-binding cell surface receptor and
   TREM2 drives the generation of lipid-associated macrophages (LAMs),
   which regulate lipid metabolism through the expression of genes such as
   Cd36, Lpl, Fabp4, Fabp5, among others ([268]74, [269]75). TREM2^+ LAMs
   display M2 macrophage phenotypes and maintain tissue homeostasis
   ([270]74–[271]76). In breast cancer, the TREM2^+ LAM subpopulation has
   been identified to support an immunosuppressive microenvironment, and
   genetic depletion of this LAM subset in mice suppressed tumor growth
   ([272]77). We speculate Trem2 loss of function in vivo may play a
   particularly important role in reprogramming LAM function in the GBM
   TME to augment the antitumor response.

   Previous reports have found that TREM2 inhibition slows the growth of
   mouse sarcoma and subcutaneous tumor, with increased response to
   anti–PD-1 treatment ([273]14, [274]15). We examined the T cell
   populations in GBM-bearing mice and found that both Trem2 knockout and
   Trem2 ASO treatment resulted in increased PD-1^+CD8^+ T cell numbers in
   GBM TME. We then tested the combined effect of Trem2 loss of function
   and anti–PD-1 treatment for GBM therapy and found that anti–PD-1
   further enhanced the therapeutic efficacy of Trem2 inhibition in
   tumor-bearing animals. We thus conclude that Trem2 inhibition
   contributes to tumor suppression in vivo by reshaping myeloid subset
   and T cell functions and that Trem2 inhibitory strategies, especially
   in combination with immune checkpoint blockade, may represent an
   attractive new approach for GBM therapy. Whether Trem2 regulates the
   phenotypes of other immune cell subsets to provide potentially
   additional therapeutic opportunities remains an open question for
   future studies.

MATERIALS AND METHODS

Study design

   The objective of this study was to determine the role of TREM2 in GBM
   tumor cell growth and evaluate whether TREM2 inhibition is of value for
   GBM therapy using both primary human GBM cells and mouse models of GBM.
   For in vitro experiments, samples were randomly selected for treatment,
   and both data collection and analysis were performed in a blind fashion
   in regard to treatment group. For in vivo experiments, mice were
   randomly assigned to treatment groups with littermate controls or
   otherwise relevant controls (control knockdown or knockout myeloid
   cells, control ASO treatment). Investigators were blinded to mouse
   genotype or group before treatments. No data were excluded from the
   analysis. Primary GBM cell cultures were generated from fresh GBM tumor
   tissue obtained from Barnes-Jewish Hospital/Washington University
   School of Medicine. Experiments with human tumor tissue were conducted
   in accordance with approved institutional review board protocols: IRB#
   201211019, 201409046. Informed consent was obtained per protocol from
   patients before usage. Experiments with animals were conducted in
   accordance with approved institutional review board protocols
   (#21-0083) and adhere to National Institutes of Health and American
   Association for Laboratory Animal Science guidelines and have been
   approved by our Institutional Animal Care and Use Committee (IACUC).

Plasmids preparation and lentiviral production

   All human and mouse gene target-directed short hairpin–mediated RNA
   (shRNA) clones in this study were purchased as bacterial glycerol
   stock, and amplificated shRNA plasmids were isolated through QIAprep
   Spin Miniprep and EndoFree Plasmid Maxi kits (QIAGEN) per instructions.
   Two human TREM2 gene RNAi with corresponding scrambled RNAi were
   selected as follows: TREM2i.1 (TRCN0000056719), scr.1 (SHC002)
   (Sigma-Aldrich), TREM2i.2 (RHS4430-200284823), and scr.2 (RHS4346)
   (GIPZ Lentiviral shRNA, Horizon Discovery). Mouse Trem2 gene RNAi with
   high knockdown efficiency were used: Trem2i.1 (TRCN0000068096) and
   Trem2i.2 (TRCN0000068097) (Sigma-Aldrich); and SHC002 was the
   nontargeting control. In this study all shRNA clones are listed as
   table S1.

   Human embryonic kidney 293T cells were used for shRNA-expressing
   lentiviral packaging. Briefly, HEK 293T cells were seeded into six-well
   plates one day before plasmids transfection. With cell confluence
   arriving at 70 to 80%, transfection was performed by making DNA/PEI
   (Polysciences) complexes into Opti-MEM medium (Gibco). The shRNA
   plasmid was mixed with packaging plasmid psPAX2 and envelope plasmid
   pCMV-VSVG by a ratio per instructions. On day 6, culture medium was
   collected, centrifuged for dead cell debris removal, and filtered
   through 0.45-μm syringe filter. Lenti-X Concentrator (Clontech) was
   then used to concentrate lentivirus in the supernatant. After
   incubation at 4°C for 6 to 7 hours, virus-containing pellets were
   collected by centrifuging at 1500g for 45 min at 4°C, resuspended in
   cold phosphate-buffered saline (PBS), and stored at −80°C in single-use
   aliquots.

Syngeneic orthotopic tumor implantation

   Two mouse syngeneic GBM models were used in this study. A total of
   10,000 cells injection per animal was performed for a slow SB28 tumor
   developing model. A total of 2000 cells injection per animal was used
   for developing NPA C54B glioma in vivo. For a coimplantation model,
   SB28 tumor cells and primary BMDMs or BV2 cells, by a number ratio of
   1:2, were mixed and coinjected into the mouse brain. Cells were
   stereotactically (David Kopf Instruments) injected into the right
   putamen of age-matched trem2 transgenic mice or 6 to 8 weeks old
   C57BL/6 mice. The coordinates used were as follows: 1.0 mm rostral to
   bregma, 1.5 mm lateral, and 2.0 mm deep. Mice’s survival was monitored
   and euthanized on the basis of neurological-deficient symptoms
   according to IACUC guidelines.

ASO design and ICV injection

   Trem2-targeted and inactive ASOs were designed as reported ([275]25)
   and synthesized (Integrated DNA Technologies). The knockdown effects of
   Trem2-lowering ASO at different concentrations were evaluated in BV2
   cells in vitro ([276]25). In vivo injection was performed as described
   ([277]25). Briefly, Trem2-targeted and inactive ASOs were diluted in
   sterile saline and administered via ICV injection. Seven days later,
   mice were stereotactically injected with GBM cells as mentioned above.
   Mice’s survival was monitored and euthanized on the basis of
   neurological-deficient symptoms.

In vivo treatments

   After tumor implantation, trem2 transgenic mice were treated with
   sotuletinib (BLZ945, brain-penetrant CSF-1R inhibitor) or vehicle (20%
   Captisol) [MedChemExpress (MCE)]. Mice were dosed with sotuletinib (200
   mg/kg) or vehicle by oral gavage at days 7, 12, and 17. Symptom-free
   survival was evaluated. Mice were treated intraperitoneally with
   anti-TREM2 antibody (clone 178; 200 μg per mouse) or isotype control
   (clone 135) at days 4, 8, 12, and 16 after tumor injection. Mice were
   intraperitoneally injected with anti–PD-1 antibody (200 μg per mouse,
   BioXCell) or isotype control (rat IgG2a) at days 8, 11, and 14 after
   tumor injection.

Generation and culture of mouse BMDMs

   The isolation and differentiation of mouse BMDMs by macrophage
   colony-stimulating factor (M-CSF) have been performed ([278]26).
   Briefly, mice between 6 and 12 weeks of age were euthanized following
   IACUC guidelines. The lower limb was exposed by removing the skin. The
   femur and tibia were collected by cutting off the base of femur into
   the spine and Achilles tendon round tibia. Major muscles and fibula
   were then removed. After removing the epiphyses of femur and tibia, BM
   cells were washed out by sterile PBS using 23-gauge needles. Cell
   pellets were collected by centrifuging at 500g for 5 min at 4°C and
   resuspended with 10 ml of Dulbecco’s modified Eagle’s medium (DMEM)
   supplemented with 10% fetal bovine serum (FBS) and 1%
   penicillin-streptomycin. Cells were filtered through a 70-μm cell
   strainer and centrifuged again. After a gentle pipetting in complete
   DMEM with M-CSF (10 ng/ml), cells were seeded into 10-cm-diameter
   polystyrene-treated tissue culture dishes for a 7-day continuous
   differentiation. On day 3, half volume of culture medium was replaced
   with fresh M-CSF–containing DMEM. On day 7, cell morphology was
   assessed, and medium was changed without M-CSF. BMDM was then prepared
   for both in vitro stimulation and in vivo injection experiments.

Primary microglia culture

   Primary microglia were isolated according to the published method
   ([279]45) with modification. Briefly, primary cortical glial cultures
   were generated from postnatal day 2 mouse pups and cultured in DMEM
   supplemented with 10% FBS. After 2 to 5 days, the media were replaced
   with fresh DMEM containing granulocyte M-CSF (2 ng/ml). Once the
   microglia were blooming, they can be purified by agitation on an
   orbital shaker at 200 rpm for 2 hours at 37°C. Cell pellets were
   collected by centrifugation at 440g for 5 min, resuspended in a 1:1
   ratio of fresh astrocyte culture media/old media, and seeded into
   24/96-well plates. The next day, the media were replaced with microglia
   treatment media supplemented with 0.1% bovine serum albumin (BSA) and
   M-CSF (10 ng/ml). Microglia from trem2 transgenic mice were assayed for
   phagocytotic activity or directly cocultured with GBM cells. Primary
   microglia (C57BL/6 background) were treated with 10 μM inactive ASO
   controls or Trem2-targeted ASO.

Phagocytosis assay

   The phagocytotic capacity of BMDMs and primary microglia from trem2
   transgenic mice and BV2 cells with Trem2 knockdown was detected using
   Fluoresbrite YG Microspheres (Polysciences). Cytochalasin D
   (Sigma-Aldrich) was used as the negative control and added into wells 1
   hour before microsphere treatment. Cells were incubated with YG beads
   (1:2000 dilution in medium) for 2 hours at 37°C. The medium was then
   removed, and cells were washed three times with 1× PBS. Cells were
   fixed with 4% paraformaldehyde (PFA; ProSciTech) at 4°C for 15 min and
   stained with 4ith stained t 4 ( with 4% pa (DAPI) (1:5000 dilution in
   PBS). Primary microglia were labeled with Iba1 (GeneTex) primary
   antibody overnight, followed by incubation in Alexa Fluor
   568–conjugated goat anti-mouse secondary antibody (1:500 dilution;
   Invitrogen) for 2 hours at room temperature. Images were captured on a
   microscope (Leica Microsystems). Bead colocalization with cells was
   quantified by ImageJ colocalization plugin and cell count (DAPI).

Flow cytometry

   SB28 GBM-bearing mouse was euthanized following IACUC guidelines, and
   the brain tumor was dissected. Tumor tissues were dissociated to
   generate single-cell suspensions using a Brain Tumor Dissociation Kit
   (Miltenyi Biotec) according to the manufacturer’s instructions. Cell
   numbers were counted and incubated with CD45 MicroBeads (Miltenyi
   Biotec) for 15 min at 4°C. Live CD45^+ cell fractions were enriched by
   running cell suspensions through MACS Columns in a Magnetic MACS
   Separator (Miltenyi Biotec). CD45^+CD11b^+ cell fractions were then
   enriched using CD11b MicroBeads (Miltenyi Biotec). Two antibody panels
   were used for flow cytometry analysis: the myeloid TME composition,
   anti-mouse fluorescein isothiocyanate (FITC)–CD45 (1:200),
   allophycocyanin (APC)-Ly6C (1:100), PerCP-CD11b (1:100) (BioLegend),
   and phycoerythrin (PE) donkey anti-rabbit immunoglobulin G (IgG)
   (1:200; BioLegend) and Trem2 polyclonal antibody (1:100; Bioss); the
   lymphoid TME component, anti-mouse APC-CD3 (1:50), PerCP-CD8 (1:100),
   and PE-CD279 (PD-1) (1:20) (BioLegend). Briefly, cells were incubated
   with TruStain FcX-Fc blocker CD16/32 (BioLegend) for 15 min at room
   temperature. Then, cells were stained with different panels of
   antibodies or isotype controls. Samples were read on LSRFortessa (BD)
   and analyzed using FlowJo_v10.8.1 software.

Hematoxylin and eosin staining

   Mouse brain with tumor implantation was fixed by 4% PFA (ProSciTech)
   perfusion and embedded into optimal cutting temperature (OCT) (Sakura)
   blocks. Frozen OCT tissue blocks were sectioned into 30 μm in thickness
   using cryostat (Leica Biosystems) and preserved at −20°C. Frozen
   sections were rehydrated in 100, 90, and 70% ethanol solution
   sequentially for 2 min at each. Tissue nuclei was stained by immersing
   sections in Gill Hematoxylin 2 (Thermo Fisher Scientific) for 3 to 4
   min. Sections were then washed in double-distilled H[2]O and dipped
   into Bluing agent (Agilent) for 30 s to complete the nuclear staining.
   After 20 dips in 95% ethanol, tissue slides were immersed in eosin
   (Agilent) for 1 min to get cytoplasm counterstain. With one dip of
   water wash, slides went through dehydration process by two times of 30
   dips in 95% ethanol solution and two 5-min washes in 100% ethanol.
   Slide tissues were mounted with Cytoseal 60 (Thermo Fisher Scientific)
   for long-term storage at room temperature.

Immunohistochemistry

   SB28 cells were stained with tracking dye: Cytopainter (green
   fluorescence) (Abcam) based on the manufacturer’s instructions. Then,
   cells were stereotactically (David Kopf Instruments) injected into the
   right putamen of C57BL/6 mice. Mouse was euthanized when
   neurological-deficient symptoms develop. Brains were dissected and
   embedded into OCT medium. Frozen OCT tissue blocks were sectioned into
   30 μm in thickness and blocked with 10% normal goat serum (Vector
   Laboratories) in PBST (PBS containing 0.1% Triton X-100) for 1 hour at
   room temperature. Sections were incubated with Iba1 (GeneTex), TMEM119
   (ABclonal), or cleaved caspase-3 (Cell Signaling Technology) primary
   antibody overnight at 4°C. After three washes in PBST, tissues were
   applied with Alexa Fluor 568/488–conjugated goat anti-mouse/rabbit IgG
   (H + L) secondary antibody (Invitrogen) for 2 hours at room
   temperature. With additional washes, sections were mounted with
   Fluoromount-G containing DAPI (Southern Biotech). Images were taken on
   a microscope (Leica Microsystems).

Conditioned medium

   Different CMs were collected for study: THP-1 macrophage-like cells
   were cultured for 48 or 72 hours, as well as under IFNγ (20 ng/ml), LPS
   (10 ng/ml), or IL-4 and IL-13 (20 ng/ml) treatment for 24 or 48 hours,
   respectively; THP-1–derived macrophages were cultured with control
   IgG1/anti-TNFα antibody (10 ng/ml; adalimumab, MCE) for 48 hours; BMDMs
   and primary microglia treated with ASOs were cultured for 48 or 72
   hours or stimulated with IFNγ (10 ng/ml) for 24 or 48 hours,
   respectively; BV2 microglia were cultured for 48 or 96 hours
   with/without IFNγ (50 ng/ml) treatment; BV2 microglia and SB28 cells
   were cocultured by a direct-contact manner. Aliquot CM was stored at
   −20°C. For cytotoxicity assays, CM was added into target cells’ medium,
   and the ratio of medium volume was 1:4.

Cytotoxicity assay

   The cytotoxic effects on GBM cells were assessed by cell
   viability/proliferation assays in this study. To measure the effects of
   CM from human THP-1–derived macrophages on human GSCs, adenosine
   5′-triphosphate (ATP) generation by viable cells was detected using
   CellTiter-Glo luminescence cell viability assay (Promega) per the
   manufacturer’s instructions, and YOYO-1 (Invitrogen) staining on dead
   cells was evaluated.
   3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfopheny
   l)-2H-tetrazolium (MTS) colorimetric reagent (Promega) was also used to
   determine the number of viable cells by measuring absorbance at 490-nm
   wavelength. Absolute cell counting and ATP assays were performed to
   examine cell viability/proliferation of SB28 cells when cocultured with
   BMDMs or BV2 microglia in a Transwell system (Corning).

Enzyme-linked immunosorbent assay

   Tumor necrosis factor (TNFα) secretion in the supernatants was assayed
   by ELISA (enzyme-linked immunosorbent assay) MAX Standard Set Mouse
   TNF-α kit (BioLegend) per the manufacturer’s instructions.

Nitrite assays (Griess)

   The NO level in cultured supernatants was measured by modified Griess
   reagent (Sigma-Aldrich). Briefly, 150 μl of supernatant per sample was
   mixed with equal volume of 1× Griess reagent in a 96-well microplate.
   After 15-min incubation at room temperature, the NO concentration
   (nitrite) was determined by measuring the absorbance at 540 nm on the
   Cytation 5 multimode reader (BioTek).

In vitro three-dimensional coculture system

   The direct interaction between microglia and GBM cells was assessed by
   three-dimensional (3D) coculture system in vitro. Briefly, primary
   microglia or BV2 cells and GBM cells were labeled with tracking dyes
   respectively: primary microglia/BV2 cells and Cytopainter (green
   fluorescence); GBM cells and Cytopainter (deep red fluorescence)
   (Abcam) based on the manufacturer’s instructions. Microglia and tumor
   cells were mixed at the ratio of 2:1 and added into the chamber slide
   (Thermo Fisher Scientific) in complete medium to support the growth of
   two cell types. Chambers were removed on days 1, 3 and 5, and cells on
   slide were fixed with 4% PFA (ProSciTech) for 15 min on ice. Slides
   were mounted with DAPI-containing medium. The supernatants on days 1, 3
   and 5 were collected and assayed for cytokines. Cells on slides were
   observed on a microscope (Leica Microsystems), and fluorescence
   intensity was quantified using ImageJ.

RNA isolation and RT-qPCR

   Cells were washed with ice-cold PBS and processed for total RNA
   isolation using QIAshredder and RNeasy Plus Mini Kit (QIAGEN) according
   to the manufacturer’s instructions. RNA was then reversely transcribed
   into cDNA by a High-Capacity cDNA Reverse Transcription Kit with RNase
   Inhibitor (Applied Biosystems). RT-qPCR was performed by using Power
   SYBR Green PCR Master Mix (Applied Biosystems) on CFX Connect Real-Time
   PCR Detection System (Bio-Rad) with a reaction setup as below: initial
   cycle, 95°C for 10 min; amplification stage, 40 cycles involving a
   denaturing step (95°C for 15 s), an annealing step (57°C for 30 s), and
   extension step (72°C for 30 s); a melting curve analysis, 65°C for 5 s
   and then 5 s each at 0.5°C increments between 65° and 95°C.
   Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as the
   internal reference gene and analysis is conducted on CFX Manager
   software (Bio-Rad) and Excel 2015 (Microsoft). Sequences for each pair
   of primers were listed in tables S2 and S3.

Western blot

   Protein levels of trem2 gene expression in primary microglia, BV2 and
   THP-1 cell samples were confirmed by Western blot. To assess
   phosphorylated states of proteins under different stimuli in primary
   microglia and BV2 cells, Western blot assays were performed as below:
   Primary microglia were treated with IFNγ (10 ng/ml) for 30 min or dead
   SB28 cells for 60 min; BV2 cells were treated with IFNγ (10 ng/ml) or
   LPS (100 ng/ml) for 30 and 60 min, respectively, or dead SB28 cells for
   90 and 120 min. Quickly, cell samples per time point were washed three
   times with cold PBS and lysed on ice for 30 min in 1×
   radioimmunoprecipitation assay buffer (Cell Signaling Technology) with
   Protease Inhibitor Cocktail (MilliporeSigma). The lysates were then
   centrifuged at 15,000g for 20 min at 4°C. The supernatants were
   collected, and concentrations were determined using Protein Assay Dye
   Reagent (Bio-Rad) per protocol. Proteins each sample were then
   separated on 10 or 12% SDS–polyacrylamide gels and transferred onto
   0.45-μm polyvinylidene difluoride membranes (Millipore). After 1 hour
   of shaking in blocking buffer (3% nonfat milk or BSA in PBS containing
   0.05% Tween 20) at room temperature, membranes were incubated with
   primary antibodies overnight at 4°C. Next day followed by three washes
   using PBST buffer (PBS with 0.05% Tween 20) for 5 min per each, the
   membrane was incubated with IRDye 800CW donkey anti-rabbit IgG or IRDye
   680RD donkey anti-mouse IgG secondary antibody (LI-COR Biosciences)
   (1:10,000 dilution in PBST buffer containing 1% BSA) for 2 hours at
   room temperature. After four washes in PBST buffer, blots on membranes
   were detected on Odyssey Fc Imaging System (LI-COR Biosciences).
   Primary antibodies used in this study include the following: TREM2
   (Proteintech); phospho-STAT1, phospho-ERK, phospho-JNK,
   phospho-IκBα-S32, STAT1, ERK1/ERK2, JNK, IκBα, TREM2, Bcl-2, and Bax
   (ABclonal); TREM-2 (B-3) and GAPDH (6C5) (Santa Cruz Biotechnology);
   phospho-p38 MAPK (D3F9) and p38 MAPK (D13E1) (Cell Signaling
   Technology).

Bulk RNA-seq and data analysis

   Primary myeloid cell populations were collected for transcriptomic
   analysis. BMDMs from trem2 transgenic mice were treated with IFNγ (10
   ng/ml), LPS (100 ng/ml), or dead SB28 cells for 6 hours. Primary
   microglia from nontransgenic mice were treated with ASOs for 3 days and
   then stimulated with IFNγ (10 ng/ml) or dead SB28 cells for 6 hours.
   Total RNA per sample was isolated using RNeasy Plus Mini Kit (QIAGEN)
   according to the manufacturer’s instructions. RNA samples were
   submitted to the McDonnell Genome Institute (Saint Louis, MO) for
   mRNA-seq library generation and sequencing (Illumina NovoSeq).
   Generated data were then analyzed according to the published method
   ([280]78). Differentially expressed genes (P[adj] < 0.05) were
   clustered using k-means clustering and heatmaps of z-scores generated
   using Phantasus. Pathway analysis for each cluster was performed using
   EnrichR-KG using the KEGG database.

3′ scRNA-seq and data processing

   Dissociated single-cell suspensions from human GBM samples (B150 and
   B152) or SB28 tumors (nontransgenic mice) were processed using the 10x
   Genomics Chromium Controller and the Chromium Single-Cell 3 oceLibrary
   & Gel Bead Kit following the manufacturer’s protocols
   ([281]https://tinyurl.com/ybpg2pfz). The yielded cDNA libraries were
   sequenced on an Illumina NovaSeq platform. All samples were processed
   using an in-house Workflow Description Language (WDL) pipeline. The
   pipeline first makes doublet predictions on raw 10x data using
   DoubletCollection wrapper, in which six doublet callers are used
   including “cxds,” “bcds,” “hybrid,” “scDblFinder,” “DoubletFinder,” and
   “doubletCells.” Doublet rate of 0.08 was used for calling these
   doublets. Data were further processed using Seurat version 4.0.0. Cells
   were filtered on the basis of nFeature > 700 (number of genes detected
   in each cell), nCount [unique molecular identifier (UMI) count] > 93rd
   percentile, and percent.mito > 0.05 (percentage of mitochondrial
   genes). Merged SB28 samples were clustered and projected on the UMAP
   using 50 principal components and a clustering resolution of 1.2. Cell
   types for clusters were predicted using Sctype. Both immune and brain
   cell markers were used in the prediction. Trem2 expressions across
   different cell clusters were ranked. Data for human GBM tumors were
   processed using our WDL pipeline as described. Cell type predictions
   were made using our in-house GBM cell type calling pipeline. Cells were
   first divided into CD45^+ and CD45^− clusters. SingleR (version 1.4.1)
   was used for reference-based cell type identification. For the CD45^+
   subset, Brain Immune Atlas ([282]56) and STAB data ([283]79) were used
   as reference datasets for cell type prediction. For the CD45^− subset,
   reference datasets included previously published ([284]18) and STAB
   data ([285]79). Cell types less than 1% abundance in clusters were
   filtered out. TREM2 expression in different cell clusters was ranked
   and plotted.

   Tumor tissues from ASO-injected mice were dissected and dissociated
   into single-cell suspensions. Live CD45^+ cell fractions were enriched
   through magnetic sorting. Mixed CD45^+ and CD45^− cell suspensions were
   submitted to the McDonnell Genome Institute (Saint Louis, MO) for 10x
   Genomics library preparation and sequencing. Unfiltered gene expression
   matrices were processed using SoupX to reduce ambient reads in each
   sample. Datasets were then further processed using Seurat v4.0. Cells
   were filtered to remove cells with % mt-reads > 5% and total genes <
   300 and > 9000 per cell. Data were then normalized using sctransform
   package with % mt-reads regressed and datasets integrated. Principal
   components analysis was performed and the first 20 dimensions were used
   in the FindNeighbors and RunUMAP commands in Seurat. Clusters were
   identified over a range of resolutions from 0.2 to 1.0. Two low-quality
   cell clusters determined by low numbers of genes and counts and
   coexpression of multiple cell type–specific markers were removed, as
   were two small clusters pertaining to astrocytes and neurons. Three
   clusters corresponding to tumor cells were also removed to focus on
   immune cells. Principal components analysis was performed again, and
   the 15 dimensions were used to replot the data. A range of resolutions
   from 0.2 to 1.0 was used to identify clusters, and data were further
   processed at a resolution of 0.4. Marker genes for each cluster were
   identified using the FindAllMarkers function by model-based analysis of
   single-cell transcriptomics (MAST) using LogNormalized RNA counts.
   Clusters were annotated according to cell type–specific gene
   expression, and the proportion of cells in each cluster for each sample
   was calculated. To perform differential gene expression analysis
   between ASO and inactive groups, pseudo-bulk matrices of summed counts
   for each cluster were constructed from n = 2 biological replicate
   scRNA-seq datasets per group. Differential gene expression for each
   cluster was then determined using the edgeR package. Pathway analysis
   of up- and down-regulated genes was performed using EnrichR-KG and the
   TRUSTT, KEGG, and Reactome databases.

The Cancer Genome Atlas data processing

   The UCSC xena browser ([286]https://xenabrowser.net/) was used to
   download relevant dataset for The Cancer Genome Atlas (TCGA) GBM
   cohort, which included survival, phenotype, and gene expression data
   profiled in the microarray HGU133 platform and the nonsilent mutation
   call data. TREM2 mRNA expression in 252 primary GBM tumor samples
   (IDH1/2–wild-type) and 10 normal brain specimens was plotted (log
   scale; horizontal line represents median, and dots indicate individual
   patient’s TREM2 expression). The Wilcoxon rank sum test was performed
   for comparison. The TCGA pan-cancer RNA-seq (v2) normalized TREM2
   expression data for low-grade glioma and GBM cohorts were downloaded
   via UCSC xena browser, including 6 normal, 65 G2 astrocytoma, 77 G2
   oligoastrocytoma, 116 G2 oligodendroglioma, 132 G3 astrocytoma, 56 G3
   oligoastrocytoma, 82 G3 oligodendroglioma, and 166 GBM patient samples.
   The Kruskal-Wallis test was used to compare TREM2 expressions across
   all groups. The PFS data of 249 patients with IDH1/2–wild-type GBM were
   further analyzed on the basis of TREM2 expression. TREM2 expression was
   dichotomized into high/low by the optimal cutoff. Kaplan-Meier curves
   were generated for high- and low-expressing patients, and log-rank test
   was performed. The OS data of patients with IDH1/2–wild-type GBM were
   queried using R package “cgdsr” via cBioPortal
   ([287]http://www.cbioportal.org/). Kaplan-Meier survival curves were
   generated for 115 patients with IDH1/2–wild-type GBM according to TREM2
   expression. Patients were divided into high and low groups based on the
   optimal cutoff of TREM2 expression, with maximized the log-rank test
   statistic as implemented in R package “maxstat.” Kaplan-Meier curves
   were shown with the number of events/total n, median survival time with
   95% confidence interval (CI), hazard ratio with 95% CI, and log-rank
   test P value.

Statistical analysis

   Significance assessments were performed with GraphPad Prism 9. A
   two-tailed Student’s t test was used for comparisons in experiments
   with two conditions. One-way analysis of variance (ANOVA) analysis was
   used for experiments with more than two groups and corrected by
   Bonferroni test for multiple comparison. Animal survival data were
   processed into Kaplan-Meier survival curves and performed log-rank
   tests. All data have been generated from at least three independent
   biological experiments unless otherwise mentioned. A P < 0.05 was
   considered significant.

Acknowledgments