Targeting TREM2 on tumor-associated macrophages enhances immunotherapy

,


In brief
Binnewies et al. show that TREM2expressing tumor-associated macrophages (TAMs) are critical mediators of immune suppression in the tumor microenvironment (TME) and correlate with T cell exhaustion in human cancer.Effector-enhanced anti-TREM2 antibody treatment alters the abundance and phenotype of TAMs in the TME and sensitizes the response to anti-PD-1 therapy.
Therapeutic targeting of TAMs and related immunosuppressive myeloid cells has so far had a minimal effect on improving clinical outcomes (Butowski et al., 2016;Calvo et al., 2017;Jahchan et al., 2019;Nywening et al., 2016).For instance, therapies that reduce the global abundance of macrophages are associated with paltry anti-tumor responses, likely because of poor tumor specificity (Jahchan et al., 2019).It is also likely that, although some TAMs are highly immunosuppressive, others are essential for tumor clearance.
Consequently, new approaches are needed to more precisely target TAMs as well as other immunosuppressive, tumor-associated myeloid cell populations (Broz et al., 2014) to improve the efficacy and safety associated with myeloid cell-targeting therapies.
Here we report transmembrane protein triggering receptor expressed on myeloid cells 2 (TREM2) as a highly promising therapeutic target because of its enriched expression on TAMs and known immunosuppressive function in human and mouse.TREM2 has been studied widely in microglia, where TREM2 functions in neuronal debris clearance to counteract the inflammatory response (Takahashi et al., 2005).Based on the known functions of TREM2 in microglial pathophysiology and recent work on TAMs in the TME, it is postulated that TREM2 on TAMs dampens inflammatory gene expression directly through its association with DNAX-activating protein of 12 kDA (DAP12) and indirectly through promotion of apoptotic cell clearance.In the present study, we found that TREM2+ TAMs corresponded to a highly immunosuppressed TME in mouse and human.Treatment of tumor-bearing animals with an Fc domain-enhanced anti-TREM2 monoclonal antibody (mAb) led to depletion of TAMs and drove anti-tumor immunity as a single agent and when combined with anti-PD-1.Further, we identified ovarian cancer as an ideal tumor indication for anti-TREM2 mAb therapy because of its poor response to CPI, high expression of TREM2, and high density of TAMs and based on the anti-tumor response to anti-TREM2 mAb in a preclinical orthotopic model of ovarian cancer.Our work not only confirms the recent findings of others but extends into a highly translationally focused, clinically relevant assessment of TREM2 on TAMs.Furthermore, we enabled not only our own findings but the findings of others by developing an anti-TREM2 mAb clinical asset that is currently being tested in humans (ClinicalTrials.gov:NCT04691375).

RESULTS
TREM2 expression is correlated with immune exhaustion and anti-PD-1 resistance in the mouse To identify genes that are significantly upregulated by TAMs in human tumor indications with a low response rate to anti-PD-1 treatment (Matulonis et al., 2019), we sorted CD45 + immune cells from a dissociated human ovarian tumor and then performed single-cell RNA sequencing (scRNA-seq).After removing dead cells from analysis, processing of 8,954 cells using Seurat (Stuart et al., 2019) led to 16 unique transcriptional clusters of myeloid and lymphoid origin (Figures 1A and 1B).Differential expression (DE) analysis between all populations identified in Figure 1A revealed genes that are enriched specifically within TAMs and identified the cell surface molecule TREM2 as being highly expressed in TAMs compared with other immune populations in the ovarian tumor sample (Figure 1C; Table S1).For simplicity, only the top 5 differentially expressed genes are displayed.This is consistent with our previous work using bulk RNA-seq, which identified TREM2 as being associated with inhibitory TAMs but not stimulatory dendritic cell populations (Broz et al., 2014).We then assessed the fine distribution of TREM2 mRNA in TAMs, monocytes, and proliferating cells (which contain some TAMs) (Figure 1D) and found that, although a small proportion of monocytes have low-level TREM2 expression, almost all TAMs have significantly higher expression.Using the monocyte and TAM clusters from the ovarian scRNA-seq experiment as input cells for cell trajectory analysis (Trapnell et al., 2014), we demonstrated a progressive increase in TREM2 expression across the pseudotime trajectory (Figure 1E), with marked decreases and increases in monocyte-and macrophage-related genes, respectively (Figure S1A).Because enriched expression of a target is key for therapeutic development, we compared the expression of TREM2, CSF1R, and CD163 in normal tissue-resident macrophages (TRMs) and TAMs in breast (Cassetta et al., 2019;Figure S1B).TREM2, compared with CSF1R or CD163, was substantially increased in TAMs compared with TRMs, indicating that TAMs are enriched for TREM2 on a per-cell basis.We confirmed a per-cell increase in TREM2 on TAMs compared with TRMs using an independent scRNA-seq dataset in colorectal cancer (CRC), stomach adenocarcinoma (STAD), and uterine corpus endometrial carcinoma (UCEC) (data not shown; Cheng et al., 2021).To extend this analysis to more indications and a larger cohort, we analyzed TREM2 expression across 9,736 tumors and 8,587 normal samples from the The Cancer Genome Atlas (TCGA) and the Genotype Tissue-Expression (GTEx) projects (Figure S1C).We observed a consistent increase in TREM2 levels in tumors compared with normal tissue, further supporting tumor-enriched expression of TREM2.TREM2 has been identified previously as a determinant of TAM phenotype in the TME (Katzenelenbogen et al., 2020;Molgora et al., 2020) and disease recurrence (Obradovic et al., 2021), and we have previously found TREM2 expression to be correlated negatively with survival (Broz et al., 2014).We evaluated how expression of TREM2 in select human cancers from the TCGA dataset correlated with genes correlated with TAMs (C1QC) and TAM polarization (MRC1 and CD163) (Figure S1D).In all comparisons, TREM2 was correlated significantly with TAM-associated genes.
To exhaustively examine the relationship of TREM2 + TAM composition and T cell composition and status in the TME, we conducted scRNA-seq on CD45 + tumor immune infiltrates from 10 different human tumors (Table S2) and aggregated the datasets to assess the transcriptional heterogeneity of myeloid cells (monocytes and macrophages), CD4 + TILs, and CD8 + TILs (Figure 2A; Figure S2A; Table S2).Cluster contribution from each sample was then assessed (Figures S2B and S2C).Aggregated sample analyses highlighted myeloid and lymphoid subclusters with unique transcriptional profiles, indicative of their phenotype or functional status, such as PDGFB-expressing (Kaneda et al., 2016) andFOLR2-expressing (Puig-Kro ¨ger et al., 2009) immunosuppressive TAMs (TAM C1 and TAM C2, respectively) and proliferating CD8 + TILs (CD8 + TIL -MKI67).Although TREM2 mRNA was observed across TAM, intermediate monocyte (Int.Mono), and monocyte clusters (Figures 2B and 2C), immunosuppressive TREM2 + TAM C1 and TAM C2 comprised approximately 30% of the total monocyte and macrophage compartment (Figure 2D).We then performed a Pearson's r correlation analysis of the relative numbers of cells belonging to each myeloid subset with those derived from the CD4 + TIL and CD8 + TIL subclusters to ascertain potential myeloid-T cell relationships across the aggregated tumor set (Figure 2E; Table S3).Strikingly, the proportion of TREM2 + TAM C1 and TAM C2 was highly correlated (0.65 and 0.39, respectively) with exhausted CD8 + TILs (CD8 + TIL -T EX ), indicating that the presence of TREM2 + immunosuppressive TAMs is indicative of an immunosuppressed and exhausted CD8 + TIL component.The correlation of TREM2 and T cell exhaustion was extended and confirmed to occur on a per-sample basis using our individual human scRNA-seq samples (Figure S2D) as well as with select indications in TCGA (Figure 2F).Although recent work has identified a TREM2 + immunosuppressive myeloid population subset in the mouse, designated Mreg (Katzenelenbogen et al., 2020), our analysis concluded that a transcriptionally equivalent population in humans is largely comprised of a mixed population of TAMs, Int.Monos, and monocytes (Figure S2E).This transcriptionally equivalent human population did not correlate strongly with exhausted CD8 + TILs, in contrast to the TREM2 + TAM C1 and TAM C2 populations identified in our work.
To extend our findings to a highly tractable system with innate PD-1 resistance, we utilized the CT26 syngeneic mouse tumor model (Efremova et al., 2018;Lau et al., 2017).Flow cytometry (Figure 3A; Figure S3E) and 10X Genomics scRNA-seq (Figures S3A-S3D; Table S4) of CD45 + immune cells from dissociated CT26 tumors demonstrated that, similarly to humans, TREM2 is specific to TAMs in this mouse TME.Additionally, we assessed the frequency of TREM2 + TAMs in the CT26 and EMT6 tumor models and found that, on average, tumors ranging between 100-300 mm 3 had less than 50% of TAMs being TREM2 + (Figure S3F).
To address how TREM2 expression changes as a function of tumor progression and TME immunosuppression, we analyzed expression of TREM2 on CD64 + F4/80 + TAMs in tumors of increasing size, finding a clear pattern of increasing TREM2 (C) Heatmap displaying the top differentially expressed genes between all clusters identified in (A).Differentially expressed genes for each cell type were filtered for those expressed in less than 10% of other cell types (Seurat percentage of cells where the feature is detected in the second group, i.e., pct.2 < 0.1).Cell types represented by multiple clusters were consolidated to simplify the analysis.(D) Violin plot for TREM2 expression in select populations from (A).Wilcoxon rank-sum test was used for determining statistical significance.(E) Monocle trajectory of monocytes and TAMs, displaying cell types (left), pseudotime (center), and TREM2 levels (right).See also Figure S1 and Table S1.levels correlating with progression (Figure 3B).To understand how tumor burden, TREM2 expression levels, and T cell exhaustion correlate, we analyzed the levels of the T cell exhaustion markers thymocyte selection associated high mobility group box (TOX) and PD-1 on CD8 + TILs in tumors of increasing size (Figure 3C).This analysis demonstrated a marked increase in TOX + PD-1 + CD8 + TILs in more advanced, immunosuppressed tumors (Figure 3D).When we treated CT26 tumor-bearing mice (Figure 3E) and assessed TREM2 surface levels on TAMs (Figure 3F) and TREM2 + TAM numbers in the TME (Figure 3G), we found that both increased with anti-PD-1 antibodies, although the magnitude did not meet statistical significance compared with control antibodies.Our data indicate that TREM2 is enriched on TAMs and that a TREM2 + TAM-rich TME is immunosuppressive and could be a contributing factor in initiating or maintaining resistance to anti-PD-1 treatment.

Fc domain-enhanced anti-TREM2 mAb treatment sensitizes anti-PD-1-resistant tumors
Based on the expression pattern of TREM2 and its putative relationship to tumor progression and CD8 + TIL exhaustion, we hypothesized that therapeutically targeting and skewing the proportion of TREM2 + TAMs through depletion to rebalance the immune infiltrate would benefit anti-tumor immune responses.To promote depletion of TREM2 + target cells, we designed two anti-TREM2 mAbs that were Fc competent (anti-TREM2-wild type [WT]) or Fc effector enhanced (anti-TREM2), differing only in their core fucosylation state (Houde et al., 2010).As expected, we did not observe any differences between the two antibodies in their ability to bind TREM2 antigen expressed on the cell surface (Figure S4A).Although fucosylated anti-TREM2-WT and afucosylated anti-TREM2 mAbs demonstrated comparable antibody-dependent cellular phagocytosis (ADCP) of TREM2 + target cells compared with isotype controls (Figure 4A; Figures S4B  and S4C), afucosylated anti-TREM2 mAbs elicited superior antibody-dependent cellular cytotoxicity (ADCC) compared with fucosylated anti-TREM2-WT (Figure 4B; Figure S4D).Anti-TREM2 induced ADCC when IL-4-elicited primary bone marrow-derived macrophages (BMDM)-expressing endogenous levels of TREM2 were used as cellular targets (Figure S4E).
To determine whether targeting of TREM2 + TAMs elicited antitumor immunity independent of anti-PD-1 treatment, we treated mice harboring anti-PD-1 sensitive syngeneic EMT6 subcutaneous tumors with the anti-TREM2 mAb (Figure 4C; Figure S4F).Compared with isotype mAb treatment, anti-TREM2 mAb treatment resulted in pronounced control of tumor growth.Because the anti-TREM2 mAb is designed to eliminate TREM2 + TAMs, we next investigated whether TAM abundance was affected following anti-TREM2 mAb treatment (Figure 4D).Anti-TREM2 mAb treatment progressively enabled a marked reduction in the proportion of TAMs, indicating that an afucosylated anti-TREM2 mAb depleted TREM2 + TAMs in the TME, consistent with the ability of anti-TREM2 mAbs to mediate ADCC in vitro.
To investigate the pharmacokinetic properties of anti-TREM2 therapy on endogenous TREM2, we analyzed TREM2 receptor occupancy (RO) and total TREM2 receptor levels on BMDMs and TAMs following in vitro (Figure S4G) and in vivo (Figure S4F) treatment with anti-TREM2.In both instances, anti-TREM2 treatment contributed to partial RO, but interpretation is complicated by the substantial increase in total TREM2 induced by anti-TREM2 therapy.

The anti-TREM2 mAb potentiates activation of intratumoral CD8 + T cells in an anti-PD-1 resistant tumor model
We next investigated whether therapeutic targeting of TREM2 + TAMs reversed anti-PD-1 resistance in the CT26 subcutaneous tumor model (Figure 5A).Although neither anti-PD-1 nor anti-TREM2 mAbs had appreciable anti-tumor activity as singleagent treatments, simultaneous anti-PD-1 and anti-TREM2 mAb treatment enabled pronounced control of tumor growth.A cohort of mice (20%-60% in different studies) was able to completely eliminate CT26 subcutaneous tumors and remain tumor free for over a month following cessation of the combination anti-PD-1 and anti-TREM2 mAb therapy.Such complete responders were rechallenged with CT26 tumor cells after a treatment-free holiday period.No palpable tumor growth occurred in these rechallenged mice, demonstrating that the anti-tumor immunity of anti-PD-1 and anti-TREM2 mAb treatment generated immune memory (Figure 5B).
Based on our findings, we hypothesized that anti-TREM2 treatment potentiates response to anti-PD-1 treatment by eliciting CD8 + TIL-driven anti-tumor immunity.To test this, we treated CT26 tumor-bearing animals with isotype mAbs, anti-PD-1, anti-TREM2, or a combination of anti-PD-1 and anti-TREM2 mAbs and then performed a series of analyses to systematically assess the microenvironment in the tumor.We first analyzed the proportion of CD8 + TILs in the TME of treated mice (Figure S5A).Consistent with our hypothesis, there was a modest but non-significant increase in CD8 + TILs observed only with combination therapy.We also quantitatively assessed CD8 + TIL infiltration by CD8a immunohistochemistry (IHC) and found that combination treatment significantly increased the numbers of CD8 + TILs in the TME (Figure S5B), with a less pronounced effect from anti-PD-1 monotherapy.To understand whether combination treatment affected the transcriptional profile of CD45 + immune cells, we performed scRNA-seq on CD45 + immune infiltrate from CT26 tumor-bearing mice treated with isotype, anti-PD-1, anti-TREM2, or anti-PD-1 and anti-TREM2 mAbs.scRNA-seq datasets from each condition were aggregated and analyzed in a manner consistent with the untreated CD45 + immune infiltrate from CT26 tumor cells shown in Figure 1.After isolating lymphoid cells from the dataset and reclustering (22,081 cells in total), we identified 14 clusters that were represented by all treatment groups (Figures S5C-S5E).Any cluster with co-expression of canonical lymphoid and myeloid genes were labeled as ''Myl/Lym combined'' and excluded from further analysis (Figure S5F).
(F) Correlation between TREM2 expression and T cell exhaustion signature in human breast cancer (BRCA), lung adenocarcinoma (LUAD), ovarian cancer (OV), and pancreatic adenocarcinoma (PAAD) samples from the TCGA dataset.See also Figure S2 and Tables S2 and S3.
Consistent with our transcriptional observations, we also found an increased abundance of interferon g (IFNg) and tumor necrosis factor alpha (TNF-a)-producing CD8 + TILs in the TME of combination-treated mice (Figure 5C).IFNg and TNF-a protein levels as well as IL-12p70 and IL-15 levels were enriched in supernatants from tumors of mAb-treated animals (Figure 5D).Because expression of individual genes can provide a selective view of a cellular state, we wanted to understand how transcription from a broader, holistic level was affected.Thus, we performed a gene set enrichment analysis (GSEA) comparing CD8 + TILs from isotype-treated and combination anti-PD-1 and anti-TREM2 mAb-treated mice.We observed that combina-tion-treated CD8 + TILs displayed a significant increase in the IFNg response pathway (normalized enrichment score, 2.128; adjusted p = 0.010) and TNF-a signaling via the nuclear factor kB (NF-kB) pathway (normalized enrichment score, 2.477; adjusted p = 0.010) (Figure 5E; Table S5).We also observed that CD8 + TILs from combination-treated mice had enrichment of glycolysis-related genes and a decrease in genes involved in oxidative phosphorylation, consistent with studies that show higher glycolysis in effector populations (Geltink et al., 2018).Additionally, GSEA comparing CD8 + TILs from anti-PD-1 and combination of anti-PD-1 and anti-TREM2 generated a complement of pathways (Table S5) similar to the prior analysis.S4.
CD8 + TILs in combination-treated mice were found to have the highest expression levels of effector T cell-related genes, such as Ifng, Ly6c1, Tbx21, and Eomes (Figure 5F; Table S6).We also observed expression of genes with a reported inhibitory role, such as Lag3, Ctla4, and Havcr2.These findings are consistent with previous observations that CD8 + TILs begin to express higher levels of activation-induced genes meant to subsequently dampen inflammation in a self-controlled manner (Agata et al., 1996;Walunas et al., 1994).These data support that anti-TREM2 treatment potentiates anti-PD-1 sensitivity and anti-PD-1-induced CD8 + TIL activation.
Anti-TREM2 mAb treatment induces profound changes in the tumor myeloid compartment Based on our findings demonstrating TAM depletion following anti-TREM2 mAb treatment and potent activation of CD8 + TILs following combination anti-TREM2 and anti-PD-1 mAB treatment, we hypothesized that combination treatment would elicit changes in the proportion and phenotype of TAMs in the mouse TME.Thus, we assessed changes in immune composition and gene expression in tumors following mAb treatment.In agreement with anti-TREM2 mAbs being ADCC competent, single-agent anti-TREM2 mAb and combination anti-TREM2 and anti-PD-1 mAb treatment resulted in a reduction of CD64 + F4/80 + TAMs and an increase in neutrophils in the TME (Figure 6A).We suspected that the effect of depletion was partially masked by infiltration of new myeloid cells, and in line with that, levels of myeloid-attracting chemokines, such as MIP-1a and MCP-1, were increased following combination treatment (Figure S6A).Upon further investigation, we found that major histocompatibility complex (MHC) class II À TAMs, which are thought to be more M2 like (Xiong et al., 2019), expressed higher levels of TREM2 (Figure 6B) and were selectively reduced by combination therapy (Figure 6C).Although a reduced proportion of M2-like TAMs would benefit anti-tumor immunity, we investigated whether there were transcriptional changes in the remaining TAMs following combination treatment.To focus our analysis on the myeloid compartment, myeloid cells (25,503 in total) identified in the scRNA-seq experiment, as described in Figure 4, were isolated and reclustered, yielding 12 clusters comprised of cells from all treatment conditions (Figures S6B-S6D).Similar to aggregation of lymphoid cells, clusters with coexpression of canonical lymphoid and myeloid cells were labeled ''Myl/Lym combined'' and excluded from further analysis (Figure S6E).We then assessed the scale and breadth of expression of known pro-and anti-inflammatory macrophage genes in TAMs across the four treatment conditions (Figure 6D; Table S7).We observed enhanced expression of pro-inflammatory genes in the residual TAMs following combination treatment.In addition to upregulation of pro-inflammatory genes, we observed increased expression of the M2-associated gene Arg1 in combination-treated TAMs, an effect seen previously by others (Gubin et al., 2018).Consistent with a more pro-inflammatory phenotype, TAMs from combination-treated tumors displayed increased expression of the co-stimulatory or activation-induced surface markers CD40 and CD86 (Figure 5E; Chen and Flies, 2013).Anti-TREM2 and anti-PD-1 mAb combination treatment induces   and S6.
pronounced effects on TAMs in the TME, including proportional changes indicative of selective depletion of M2-like TAMs as well as phenotypic alterations in the residual TAMs toward a pro-inflammatory phenotype.

Ovarian cancer is TREM2 rich and may benefit from anti-TREM2 mAb therapeutic intervention
To understand the scale and distribution of TREM2 surface expression on myeloid cells in the human TME, we performed flow cytometric analysis on 49 dissociated human tumor samples from distinct cancer indications (Figure 7A; Figure S7A).Extending and confirming our scRNA-seq analyses, TREM2 protein was expressed primarily on TAMs compared with all other analyzed myeloid populations.To identify indications that are TREM2-and TAM-rich, we extended this analysis by comparing the scale of TREM2 surface levels on TAMs and the proportion of TAMs in the TME (Figure 7B).Of all analyzed indications, ovarian cancer was the most TREM2 and TAM rich, with 4 of 9 tumor samples having a TREM2 geometric mean fluorescence intensity (gMFI) above 1,000 and TAMs representing more than 10% of the CD45 + immune infiltrate.
Ovarian cancer has so far demonstrated a modest response to CPI therapy (Matulonis et al., 2019).To understand how TREM2 + TAMs change with progression of ovarian cancer, we used IHC to analyze TREM2 expression in human ovarian cancer with increasing stages of disease severity (Figure 7C; Figure S7B).We found that, although a TREM2 H-score (a cumulative measure of receptor frequency and staining intensity) was low on normal ovaries, the TREM2 H-score increased steadily beginning at stage 1 through stage 3 of ovarian cancer.This phenomenon was also largely confirmed in liver and colon cancer (Figure S7C).Next, to address the relationship of TREM2 expression in individuals with ovarian cancer and disease outcome, we utilized a dataset that contained gene expression and recurrence-free survival from 285 individuals with ovarian cancer (Tothill et al., 2008).Analysis of upper-and bottom-quartile TREM2 gene expression and recurrence-free survival from 285 individuals with ovarian cancer (Tothill et al., 2008) showed worse recurrence-free survival of individuals with the highest TREM2 expression (Figure 7D).These data are consistent with TREM2 + TAMs being central mediators in driving an immunosuppressed TME as disease grade worsens.
To extend our findings to a tractable model, we tested whether the orthotopic syngeneic ovarian cancer model, ID8 Luc2 , would respond to anti-TREM2 mAb treatment (Figures 7E and 7F).Although isotype mAb-treated mice had progressive disease, mice treated with anti-TREM2 mAbs demonstrated a reduced tumor burden.These data indicate that individuals with a highly immunosuppressive TREM2 + TAM-rich TME, such as those with ovarian cancer, could benefit from therapy with an antibody that targets TREM2 + -expressing TAMs.S7.

DISCUSSION
Here we described TREM2 as a TAM-enriched target in CPIresistant human and mouse TMEs and demonstrate that therapeutic targeting of TREM2 with Fc domain effector-enhanced antibodies can drive productive antitumor immunity in tumors that are otherwise resistant to CPI therapy.Our work is relevant to current and future clinical efforts focused on identifying orthogonal therapeutic modalities that will synergize, enhance, and enable existing CPI therapies.
Although the relationship between TREM2 and immunosuppression in the context of the TME has been investigated previously (Katzenelenbogen et al., 2020;Molgora et al., 2020), much of what is known about TREM2 and its role in inflammation has been drawn from studies of microglia and the central nervous system.TREM2 in microglia promotes phagocytosis of apoptotic neurons, and, through its association with DAP12, it can downregulate transcription of pro-inflammatory genes like Tnfa, Il1b, and Nos2 (Takahashi et al., 2005).By extending this to the context of the TME, where apoptotic cell products are common and TAMs are abundant, it seems likely that TREM2 would promote a chronic immunosuppressive state in TAMs, in turn contributing to a broadly immunosuppressive immune milieu and ultimately driving CPI resistance.
In the case of human cancer, where access to tissues is a limiting factor, scRNA-seq can be utilized for its unbiased and high-dimensional assessment of broad cell-type-specific expression analysis.With a relatively small cell number input, we were able to clearly demonstrate TREM2 as being human TAM enriched.Although TREM2 expression has been noted previously on human TAMs (Katzenelenbogen et al., 2020;Lavin et al., 2017;Molgora et al., 2020), our analyses enabled highly granular assessments of the distribution of TREM2 expression in myeloid populations and the effect of TREM2 + TAMs on the prevalence of activated or exhausted T cells in the human TME.
This correlative evidence from human tumor samples was complemented by functional preclinical evidence of TREM2mediated immunosuppression in tumors using the anti-PD-1resistant mouse tumor model CT26.We found that disease severity correlated with an increase in TREM2 expression, which corresponded to more pronounced T cell exhaustion.Failed anti-PD-1 therapy elicited higher surface TREM2 on TAMs and a trending increase in TREM2 + TAMs, highlighting the potential role of TREM2 and TREM2 + TAMs in immunosuppression and promotion of tumor growth.
Therapeutic reduction of TAMs through blockade of monocyte or macrophage chemoattraction in the case of the CCL2 and CCR2 axis or differentiation with CSF-1 and CSF-1R has been attempted in pre-clinical and clinical settings with unproven success (Jahchan et al., 2019).We reasoned that the restricted expression profile on TAMs and putative role of TREM2 in immunosuppression would make it an optimal target for development of afucosylated, effector-enhanced antibodies that could elicit ADCC-or ADCP-mediated TAM elimination.N-linked glycosylation in the Fc region of antibodies has a profound effect on the capacity of an antibody to mediate ADCC through its interaction with FcgR on effector populations.Through use of glycoengineering, minimization or abrogation of fucose on Fc N-glycans can yield antibodies with improved capability to elicit ADCC at relatively lower antibody concentrations (Yamane-Ohnuki and Satoh, 2009).Consistent with the work of others (Shields et al., 2002), our core afucosylated Fc domain effector-enhanced anti-TREM2 mAb outperformed the anti-TREM2 fully fucosylated mAb in ADCC and ADCP, enabling us to perform paired efficacy and pharmacodynamic testing with the anti-TREM2 mAb as a single agent, where we observed pronounced antitumor activity and depletion of TAMs in the TME.
Our data and prior work suggested that TREM2 functions as an innate immune resistance mechanism in response to failed anti-PD-1 therapy.To understand whether anti-TREM2 mAb therapy might enhance the response to anti-PD-1, we employed the anti-PD-1-resistant CT26 tumor model.We only observed efficacy when anti-PD-1 and anti-TREM2 mAbs were combined and, thus, hypothesized that anti-TREM2 mAb therapy potentiated anti-PD-1 response through enhanced T cell infiltration or activation and modulation of TAM composition.In the pharmacodynamic analysis, we observed that CD8+ T cells were more abundant within the TME and that they produced higher amounts of effector cytokines and that their activation state, determined through transcriptional analysis, was improved.Consistent with anti-TREM2 mAb being an effector-enhanced antibody, we found that productive anti-TREM2 mAb therapy elicited a significant reduction in TAMs in the EMT6 and CT26 tumor models.In no instance did we observe a total reduction in TAMs, suggesting that, in addition to depletion, there may be alternative mechanisms of action that benefit antitumor immunity.Although the overall decrease in TAMs was moderate, we believe that production of myeloid-attracting chemokines (Figure 6) increases de novo infiltration and differentiation of M1like TAMs, as evidenced by the increases in abundance of MHC class II + TAMs and enhanced pro-inflammatory gene expression.The absence of robust depletion as a single agent in the CT26 model requires further analysis, but we suspect that it involves inadequate ADCC-or ADCP-capable effector populations when treated with anti-TREM2 mAbs alone.Furthermore, our data cannot exclude the possibility that our antibody blocks the normal immunosuppressive function of TREM2 and, in doing so, potentiates antitumor immunity.
Previous work using IHC has demonstrated that TREM2 + cells are abundant and highly prevalent in the human TME (Molgora et al., 2020).To expand these findings, we performed multiparametric flow cytometry on dissociated tumor immune cells from 50 human tumor samples in which we assessed TREM2 expression across common myeloid populations.As expected, based on our scRNA-seq data, TAMs had higher TREM2 surface levels than the other analyzed myeloid populations across all tested tumor types.Because our data demonstrated that ovarian cancer had the largest average proportion of TREM2 + TAMs within our flow cytometry dataset, we used this indication for a more detailed analysis.We observed that, as ovarian cancer grade becomes more severe, a marked increase in TREM2 levels is detectable via IHC.This grade-specific increase in TREM2 was also observed in liver cancer and, to a lesser extent, in colon cancer.Consistent with the relationship between TREM2 + TAMs and disease grade severity, TREM2 expression levels were correlated inversely with recurrencefree survival in ovarian cancer.A central feature of the CPI resistance in ovarian cancer might be related a highly immunosuppressed TME as a result of abundant of TREM2 + TAMs.To determine whether anti-TREM2 mAb therapeutic intervention might benefit ovarian cancer, we employed the ID8 Luc2 orthotopic ovarian cancer model.Strikingly, as single-agent therapy, the anti-TREM2 mAb was able to drive a robust antitumor response in this model.These data highlight not only the critical suppressive axis TREM2 + TAMs promote but also the clinical value of therapeutically targeting TREM2 to improve anti-tumor immune responses in individuals with cancer.To assess the therapeutic value of anti-TREM2 mAb-mediated therapy, we developed a humanized anti-TREM2 mAb, PY314, that is currently being tested clinically tested in individuals with solid tumors (ClinicalTrials.gov:NCT04691375).
sequencing expression data of 9,736 tumors and 8,587 normal samples from the TCGA and the GTEx projects, using a standard processing pipeline and is developed in the Zhang Lab at Peking University (Tang et al., 2019).

Mouse tumor dissociation and flow cytometry staining
Mouse tumor tissue was harvested and placed in ice cold RPMI-1640 (Invitrogen) media.Tumors were enzymatically dissociated using the Mouse Tumor Dissociation Kit (Miltenyi) according to the manufacturer's recommendation.Following dissociation, single cell suspensions were pelleted and tumor supernatant was collected, spun at high speed to remove insoluble material, enzymatically inactivated using Halt Protease Inhibitor Cocktail (Thermo Scientific) and promptly frozen at À80 C until downstream analysis was performed.Cell pellets were resuspended in stain media (DPBS/1% BSA/2 mM EDTA) and passed through a 100 uM filter to remove undissociated material.Single cell suspensions were counted on a ViCell XR (Beckman Coulter) and plated in 96-well V bottom plates for flow cytometric staining.
Cells were incubated with Zombie NIR (BioLegend), followed by FcgR block using a combination cocktail of TruStain FcX PLUS (Biolegend), Mouse Serum, Serum, Hamster Serum (Jackson Immuno Research), all prepared in Fc Receptor Blocker (Innovex).Cell surface proteins were stained for 30 minutes on ice, followed by either a secondary stain step or fixation with 1% PFA overnight at 4 C.For staining intracellular proteins, cells were fixed and permeabilized with the FoxP3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific).Intracellular antibodies were prepared in permeabilization buffer with 2% rat serum and cells were incubated for at least 30 minutes at room temperature.Cells were run on an Attune NXT (ThermoFisher).Flow cytometric analysis was performed using FlowJo (Beckton Dickinson)

T cell restimulation
Single-cell suspensions from dissociated tumors were cultured in complete RPMI-1640 media with 50 ng/ml PMA (Sigma-Aldrich), 500 ng/ml ionomycin (Thermo Fisher Scientific), and 1X Protein Transport Inhibitor (BD Biosciences) for 5-6 hours at 37 C, 5% CO 2 .Following incubation, stimulated cells were then antibody stained for surface and intracellular proteins as described above.

Human tumor and flow cytometry staining
Human dissociated tumor cells from Discovery Life Science were thawed in a 37 C water bath and freezing buffer was diluted by slowly adding 37 C media (RPMI with 10% FBS).Cells were pelleted at 400G, washed once with cold PBS, and counted on a ViCell XR (Beckman Coulter).
Cells were incubated with Live/Dead Fixable Aqua (Invitrogen), followed by FcgR block using a combination cocktail of Human TruStain FcX and True-Stain Monocyte Blocker (Biolegend), Mouse Serum, Rat Serum, Hamster Serum, and Human Serum (Jackson Immuno Research), all prepared in Fc Receptor Blocker (Innovex).Cell surface proteins were stained for 20 minutes on ice.For staining of intracellular proteins, cells were fixed and permeabilized with the FoxP3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific) per the manufacturer's instructions.Intracellular blocking was done for 20 minutes using a cocktail of Human TruStain FcX and Tru-Stain Monocyte Blocker (Biolegend), Mouse Serum, Rat Serum, Hamster Serum, and Human Serum (Jackson Immuno Research), all prepared in permeabilization buffer.Intracellular antibodies were prepared in permeabilization buffer, added to cells in blocking buffer, and incubated for 20 minutes at room temperature.Cells were washed in stain media and run on a BD FACSAriaFusion (Beckton Dickinson).Flow cytometric analysis was performed using FlowJo (Beckton Dickinson)

Antibody generation
Anti-TREM2 mAb development was performed as previously described in the WIPO patent application WO2020123664A1.Anti-TREM2 antibodies (fucosylated Anti-TREM2-WT mAb and afucosylated anti-TREM2 mAb) were expressed recombinantly with in Expi293F cells and Expi293F FUT8 knock out cells (kindly provided by University of Toronto).The gene fragments encoding the heavy and light chain residues were codon-optimized for human cell expression.The antibodies were purified from supernatant by Mab-Select PrismA column (GE Healthcare cat# 17549854) using standard protocol.After purification, the antibodies were concentrated and exchanged into PBS.

Mouse FcgRIV (ADCC) reporter assay
To perform the FcgRIV reporter assay we used the Mouse FcgRIV ADCC Bioassay (Promega).HEK293T cells transduced with murine TREM2 or murine DAP12 were used as targets and the provided Jurkat/FcgRIV/NFAT-Luc cells provided with the kit were used as effector cells.25,000 target cells resuspended in RPMI + 10% Fetal Bovine Serum and b-mercaptoethanol were added to white wall 96-well plates (Corning, 25ml/well).Anti-TREM2 or isotype mAb was prepared in RPMI + 10% Fetal Bovine Serum and b-mercaptoethanol and added to plates containing target cells followed by incubation for 30 minutes at 37 C. Single-use vials of Jurkat/FcgRIV/NFAT-Luc cells were resuspended in RPMI+10% Fetal Bovine Serum and b-mercaptoethanol and were seeded into white wall 96-well plate (25ml/well, the effector: target ratio was 3:1), containing the pre-incubated target cell: antibody mixture (50 ml/well), followed by incubation at 37 C for 5 hours.Luciferase activity was measured by using a ONE-Glo Luciferase Assay Reagent (Promega) and Tecan Spark plate reader (Tecan).The luciferase activity was normalized to the negative control, and the percentage of activity was plotted against the concentration of anti-TREM2 mAbs.The EC 50 value of each mAb was calculated by 4-parameter non-linear regression.
For the ADCP assay, IFN-g/LPS-induced BMDM served as effector cells and GFP+ HEK293T cells transduced with murine TREM2 or murine DAP12 were targets.After harvest, effector BMDM were stained with Cell-Trace Violet dye (Invitrogen) for 20 minutes at 37 C. 50,000 target cells were plated in 96-well U-bottom plates and co-incubated with anti-TREM2 or isotype mAb serially diluted prepared in media for 30 minutes at 37 C. Effector cells were then added to Antibody-Target plates at a 3:1 ratio (150,000 effectors to 50,000 targets) and incubated overnight (18 hours) at 37 C. Following incubation, cells were viability stained using Zombie NIR (Bio-Legend) then fixed using 2% Paraformaldehyde (Invitrogen) for 20 minutes at room temperature prior to being run on an Attune NXT flow cytometer (ThermoFisher).ADCP activity was measured using bead-calculated absolute counts of Cell Trace Violet-, GFP+ cells.

Cytokine analysis
Mouse tumor supernatant samples were evaluated for cytokine levels using the V-PLEX mouse cytokine 19-plex kit from Meso Scale Discovery (MSD, Cat.No. K15255D).This product contained two multiplex panels: proinflammatory panel (10 analytes) and cytokine panel (9 analytes).The MSD multiplex assay plates were precoated with capture antibodies.Samples for analysis or kit standards were added at a volume of 50 ml per well after pre-diluting the original sample with assay diluent.The plates were washed after a two-hour incubation at room temperature with agitations.Sulfo-tagged detection antibodies were added and incubated for another two hours at room temperature with agitations.Following the incubation, plates were washed once again.2X Read Substrate was added and plates were read on MSD reader.All data were analyzed by MSD Discovery Workbenchâ Software 4.0.

FFPE tissue acquisition
The ovarian, liver, and colon cancer microarrays (TMA) was purchased from Reveal Biosciences (San Diego, CA) and included patient cases in duplicate cores (2mm 2 ) with different diagnoses (Pathology, Grade, and TNM stage).The TMA at Reveal Biosciences was made by acquiring tissues that were fixed in 10% neutral buffer formalin for 24 hours and processed using identical SOPs.Sections were picked onto Superfrost Plus or Startfrost Adhesive slides and all TMAs were cut fresh in 4um serial sections upon ordering and stored at 4 C prior to IHC staining.

CD8a IHC staining
Histology was performed by HistoWiz Inc. based on their institutional SOP and fully automated workflow.All tumor samples (n = 5-6 from each of the 4 treatment groups) were processed, embedded in paraffin, and cut into 4 mm thin sections.Immunohistochemistry was performed using a Bond Rx autostainer (Leica Biosystems) with Heat-Induced Epitope Retrieval (HIER) at pH 9.0 for 20 minutes.Antibodies used were biotin conjugated rat anti-mouse monoclonal CD8a primary antibody (clone 4SM15; eBioscience diluted for use at 1:100 or 5mg/ml) and the Ready-To-Use Novocastra Streptavidin-HRP Detection System by Leica.Bond Polymer Refine Detection (Leica Biosystems) was used according to manufacturer's protocol.After staining, sections were dehydrated and covered using a TissueTek-Prisma and Coverslipper (Sakura).Each slide was scanned at 40x magnification using the Aperio AT2 scanner.Approximately 6 areas delineating the whole tumor were captured at 10x and quantified using ImageJ software (https://imagej.nih.gov).
ImageJ was used to quantify the total count of CD8a expressing cells in each section and expressed as percentage threshold area (percent of pixel in each image) of the section.All images were taken as snapshots under identical conditions such as scales, magnification and similar region of interest.Images were processed by RGB image splitting and blue channel was selected to obtain maximum separation of CD8a signals from the background tissue.Threshold setting was set between 0 and 110 for all tumors and macro format was used to automate for the batch analysis.Pixel size was selected according to the tissue morphology to exclude false positive staining.Pixel size was selected according to tissue morphology in order to exclude false positive staining.Circularity setting was between 0 and 1 for this quantification in order to capture both non-circular and circular signals in the analysis.The percentage of area and total number of CD8a counts were measured by ImageJ.For analysis, the cell count average of all images was plotted as mean ± the SD to look at CD8a expression in the tumors of 6 animals (within the mean tumor weight) within each treated group.Dunnett's Multiple Comparisons test was used as the statistical method, with the Isotype serving as the control group comparator.The difference between groups was considered statistically significant when p values were equal or less than 0.05.

TREM2 IHC staining
The IHC assay to detect TREM2 positive cells in formalin fixed paraffin embedded (FFPE) tissues was performed according to the steps described here.Slides were first baked in a 60 C oven for 45 minutes followed by deparaffinization three times in xylenes for 5 minutes each.Slides were then rehydrated in a series of ethanol gradient, from 100% to 70% ethanol, and washed at the end with distilled water.The declocker pressure cooker (Biocare) was used for the heat-induced antigen retrieval step in sodium citrate buffer at pH 6.0 (Sigma, C9999), for 15 minutes at 110 C. To block endogenous peroxidase activity, a blocker solution (Vector Labs) was applied on the slides for 15 minutes followed by rinsing in PBS-T buffer (Alfa Aesar-J63596).Non-specific binding was blocked by incubating the tissue sections with a blocking solution containing goat serum (Vector Labs) overnight at 4 C. PIT2D, also known as the recombinant anti-TREM2 antibody clone EPR20243 (Abcam, ab209814) was used as the primary antibody for 60 minutes at Room Temperature, at a concentration of 5ug/ml in PBS.Slides were then washed twice in PBS-T for 5 minutes each, followed by a 20 minutes incubation in HRP-polymer conjugated anti-rabbit secondary antibody at 1:500 dilution (MP-7500 detection kit from Vector Labs).Slides were washed twice with PBS-T for 5 minutes each before proceeding to the detection step.The DAB substrate was prepared according to the manufacturer's instructions (Abcam, ab64238) and applied to the slides for 3 minutes, followed by rinsing thoroughly in distilled water.Hematoxylin was used to counterstain the slides for 30 s followed by rinsing in running water, dehydration in series of ethanol gradient from 70% to 100% ethanol, and drying in xylenes.The stained slides were finally mounted in media and topped with coverslips for overnight drying.

Imaging and scoring
After whole slide scanning at 40x using the Aperio AT2 Scanner, quantification and intensity of the TREM2+ cells were assessed by a board-certified pathologist using the following scoring system: 0: no staining in the stromal area within the core, 1: $25% of positive cells in the stromal area within the core, 2: 50% of positive cells in the stromal area within the core, 3: 75% of positive cells in in the stromal area within the core.Percentages of positive cells that fell between these groups were scored as 0.5 ($12.5%),1.5 ($37.5%),2.5 ($62.5%), and 3.5 ($90%).The intensity of the TREM2 staining was determined as low (score of 1), moderate (score of 2), and strong (score of 3).CD163 staining was used as the positive control to identify the macrophages in the stroma of tumors and normal cores in the serial sections of the TMA (data not shown).
Lost or folded cores with more than half of the area distorted were removed from the analysis and not scored.The H-scores were calculated by using the percentage of cells (0, 12.5, 25, 37.5, 50, 62.5, 75, 90%) with intensity of each markers expression on a fourpoint semiquantitative scale, this being 0 (null, negative), 1+ (low or weak staining), 2+ (medium or moderate staining), and 3+ (high or strong staining).Thus, scores range from 0 to 300.H-scores were then calculated using the formula: H À Score = ½ð% at < 1Þ 3 0 + ½ð% at 1 + Þ 3 1 + ½ð% at 2 + Þ 3 2 + ½ð% at 3 + Þ 3 3 In the case of TREM2 staining, the intensity was homogeneous in each core and the H-score was therefore calculated by multiplying the frequency of the staining (% positive cells) with the single intensity score.

QUANTIFICATION AND STATISTICAL ANALYSIS
Unless explicitly stated, data is from a representative experiment of R 2 independent experiments.Experimental group assignment was determined through randomization.Unless otherwise noted, error bars represent ± SEM calculated using Prism.Apart from bioinformatic work, all Statistical analyses were performed using GraphPad Prism software.Unless noted otherwise, unpaired t tests were used for pairwise comparisons.For statistical measures between more than two groups, one-way ANOVA with multiple comparisons against a control group would be performed unless otherwise noted.Comparisons that did not reach statistical significance are not shown.Investigators were not blinded to group assignment during experimental procedures or analysis.

ADDITIONAL RESOURCES
A clinical trial sponsored by Pionyr Immunotherapeutics studying the use of anti-human TREM2 (PY314) as a treatment for patients with advanced solid tumors is ongoing.The clinical trial number is NCT04691375 (https://clinicaltrials.gov/ct2/show/NCT04691375).

Figure 1 .
Figure 1.TREM2 is expressed highly and specifically in human TAMs (A) Uniform manifold approximation and projection (UMAP) plot and graph-based clustering of human CD45 + tumor immune infiltrate from an ovarian tumor.(B) Dot plot of cell type-identifying genes in identified cluster types.Cell types represented by multiple clusters were consolidated to simplify the plot (e.g., monocytes).(C) Heatmap displaying the top differentially expressed genes between all clusters identified in (A).Differentially expressed genes for each cell type were filtered for those expressed in less than 10% of other cell types (Seurat percentage of cells where the feature is detected in the second group, i.e., pct.2 < 0.1).Cell types represented by multiple clusters were consolidated to simplify the analysis.(D) Violin plot for TREM2 expression in select populations from (A).Wilcoxon rank-sum test was used for determining statistical significance.(E) Monocle trajectory of monocytes and TAMs, displaying cell types (left), pseudotime (center), and TREM2 levels (right).See also Figure S1 and TableS1.

Figure 2 .
Figure 2. TREM2 + immunosuppressive TAMs are correlated with an exhausted T cell state in the human TME (A) UMAP plots of monocytes and macrophages (left) or CD4 + /CD8 + TILs (right) from 10 individuals' aggregated CD45 + immune infiltrate.(B) Violin plots displaying expression probability differences for TREM2 in monocyte and macrophage clusters from (A). (C) Frequency of TREM2 + myeloid cells in transcriptionally distinct clusters.(D) Proportion of monocytes and macrophages comprised of each cluster from 10,000 randomly sampled cells from (A). (E) Heatmap of Pearson correlation R values generated from the cell sample fractions in each cluster comparison (relates to A). Pearson correlation R values were constructed by comparing the per-tumor number of cells derived from each myeloid subcluster with all CD4 + TIL and CD8 + TIL counterparts.Each tile in the heatmap is colored by the resultant correlation value across all aggregated tumors.Significance for each comparison was computed via Monte Carlo permutation.

Figure 3 .
Figure 3. TREM2 expression correlates with tumor size and anti-PD-1 resistance in mouse (A) Representative histograms from flow cytometry data of TREM2 levels on intratumoral myeloid (left) and lymphoid cells (right).(B) TREM2 surface levels on TAMs at different tumor volumes (left) with representative histograms (right) (n = 6/group).(C) Representative histograms of intracellular TOX (left) and surface PD-1 (right) on CD8 + TILs at different tumor volumes.(D) Frequency of CD8 + TILs that are PD-1 + TOX + at different tumor volumes (n = 5/group).X-axes as seen in (B).(E) Tumor growth from CT26 tumor-bearing mice treated with isotype (left) or anti-PD-1 (right).Vertical dotted lines indicate days of treatment.(F) TREM2 surface levels on TAMs 2 days after the second dose of isotype or anti-PD-1 (left) and representative histograms (right).Unpaired t test was performed.(G) The absolute number normalized to tissue weight of TREM2 + TAMs 2 days after the second dose of isotype or anti-PD-1.Unless specified otherwise, all experiments were performed two or more times.For graphs with error bars, mean ± SEM is shown.*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.See also Figure S3 and TableS4.

Figure 7 .
Figure 7. OV is TREM2 rich and might benefit from anti-TREM2 mAb therapeutic intervention (A and B) Flow cytometry analysis of the human TME (n = 49) detailing TREM2 surface expression (delta gMFI = gMFI TREM À gMFI Isotype ) on myeloid populations with group mean value displayed (A) or comparing TREM2 delta gMFI with TAM density as a frequency of CD45 + immune cells (B).For statistical comparison in (A), Dunn's multiple comparisons test was performed with individual groups tested versus TAMs.BLCA, bladder cancer; CRC, colorectal cancer; EC, endometrial cancer; HNC, head and neck cancer; KID, kidney cancer; PCA, prostate cancer.(C) TREM2 IHC H-score from normal human ovarian tissue, benign ovarian growths, and OV grades I, II, or III.(D) Analysis of an OV dataset that contained expression data and recurrence-free survival data.Individuals in the upper and lower quartile of TREM2 mRNA expression were identified (left) and assessed for recurrence-free survival (right).(E) Experimental schematic for (F).(F) C57BL/6 mice were implanted orthotopically with ID8-Luc cells and subsequently treated with isotype or anti-TREM2 mAb (n = 10/group).Vertical dotted lines indicate days of mAb treatment.Comparisons were made using Mann-Whitney U test.Representative luminescence images of isotype and anti-TREM2 mAb treated ID8 Luc2 tumor bearing mice are shown.For graphs with error bars, mean ± SEM is shown.Unless specified otherwise, all experiments were performed two or more times.*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.See also Figure S7.