Small-molecule toosendanin reverses macrophage-mediated immunosuppression to overcome glioblastoma resistance to immunotherapy

Abstract

T cell–based immunotherapy holds promise for treating solid tumors, but its therapeutic efficacy is limited by intratumoral immune suppression. This immune suppressive tumor microenvironment is largely driven by tumor-associated myeloid cells, including macrophages. Here, we report that toosendanin (TSN), a small-molecule compound, reprograms macrophages to enforce antitumor immunity in glioblastoma (GBM) in mouse models. Our functional screen of genetically probed macrophages with a chemical library identifies that TSN reverses macrophage-mediated tumor immunosuppression, leading to enhanced T cell infiltration, activation, and reduced exhaustion. Chemoproteomic and structural analyses revealed that TSN interacts with Hck and Lyn to abrogate suppressive macrophage immunity. In addition, a combination of immune checkpoint blockade and TSN therapy induced regression of syngeneic GBM tumors in mice. Furthermore, TSN treatment sensitized GBM to Egfrviii chimeric antigen receptor (CAR) T cell therapy. These findings suggest that TSN may serve as a therapeutic compound that blocks tumor immunosuppression and circumvents tumor resistance to T cell–based immunotherapy in GBM and other solid tumors that warrants further investigation.

INTRODUCTION

Immunotherapies that activate or use T cells by immune checkpoint blockade or adoptive T cell transfer have yielded exceptional efficacy in a subset of human cancers. However, these treatments have yet to exhibit clinical activity in many of the most common and devastating solid tumors (1–3). In particular, immunologically inert tumors, such as those characterized by a paucity of tumor T cell infiltrates, such as glioblastoma (GBM), exhibit prominent resistance to T cell–based immunotherapy (4–7). This is largely due to a lack of universal, tumor-specific antigen targets and an immune-hostile microenvironment that limits T cell infiltration into and activation at the tumor. Development of new therapeutic agents that can reverse protumor immunity and overcome the microenvironment-mediated tumor resistance to immunotherapy is, therefore, desperately needed for treating solid tumors.

GBM is the most common and aggressive primary malignant brain tumor in adults, with a median overall survival of about 14 to 20 months (8, 9). GBM remains incurable and inevitably recurs after conventional cytotoxic treatments including radiation and chemotherapy (10–14). Consistent with its immunologically cold nature, GBM is also refractory to T cell–based immunotherapies including programmed cell death protein 1 (PD1)/programmed cell death ligand 1 (PDL1)/cytotoxic T-lymphocyte–associated protein 4 (CTLA4) targeting checkpoint inhibition and adoptive cell transfer with chimeric antigen receptor (CAR)–modified T cells, largely due to an immunosuppressive microenvironment that abrogates T cell activity (4, 5, 15, 16). GBM is featured by extensive infiltration with myeloid cells, such as macrophages that make up as much as half of the non-neoplastic cells (17). In the tumor microenvironment, macrophages acquire robust immunosuppressive phenotypes through genetic reprogramming and functional polarization, which blocks antitumor immunity of cytotoxic T cells by production of immunosuppressive molecules and by direct cell interaction (18–21). Likewise, recent work from our laboratory and others showed that tumor-associated macrophages drive cancer progression, immunosuppression, and therapeutic resistance in GBM (22–26), therefore representing a vital target for GBM immunotherapy. In this study, we found a small-molecule compound that can reverse macrophage-mediated immunosuppression in GBM, providing a unique opportunity to improve immune checkpoint blockade and CAR T immunotherapy in solid tumors.

RESULTS

Toosendanin is a small-molecule compound that blocks macrophage-mediated immunosuppression

We performed a functional screen to identify small-molecule compounds that can reverse macrophage-mediated immunosuppression in an in vitro tumor microenvironment system. Human peripheral blood mononuclear cell (PBMC)–derived macrophages were lentivirally transduced to express genetic probes by which the expression of firefly and renilla luciferases (fLuc and rLuc) was controlled by the promoters of interleukin-10 (IL-10), a major immunosuppressive cytokine in cancer, and cytomegalovirus (CMV; as a control), respectively. This was followed by treatment with a compound library and incubation with tumor-conditioned medium and human GBM tumor–derived endothelial cells (ECs) that have robust capacity to induce immunosuppressive macrophage phenotypes (22). Our initial screen with a customized natural product library identified several candidate compounds that inhibited IL-10 transcription in the tumor-educated macrophages (Fig. 1, A and B). Among them, toosendanin (TSN) was the top candidate in the list, leading to dose-dependent inhibition of IL-10 transcription in tumor-educated macrophages with minimal effects on basal viability in human macrophages, astrocytes, T cells, and glioma cells (Fig. 1, B to D, and fig. S1). These results were confirmed by flow cytometry analysis of tumor-educated macrophages, showing that TSN inhibited expression of CD206, a surface marker of immuno-suppressive alternatively polarized M2-like macrophages, as well as of IL-10 (Fig. 1, E and F). Moreover, TSN at 10 nM almost completely abrogated CD206 and IL-10 expression in Mϕs treated with IL-4 and IL-6 (Fig. 1, G and H), two cytokines known to induce macrophage immunosuppressive polarization. Furthermore, TSN enhanced carboxyfluorescein diacetate succinimidyl ester (CFSE) diffusion into the proliferative T cells (Fig. 1, I to K), increased CD25⁺CD3⁺ activated T cells, and up-regulated expression of effector cytokine interferon-γ (IFN-γ) and proliferative marker Ki67 in CD8⁺ T cells (Fig. 1, L to N). These results suggest that TSN rescued T cell proliferation and reactivated T cells in the presence of immunosuppressive, tumor-educated macrophages. Together, these data identify TSN as a small-molecule immunomodulator that reverses macrophage-mediated T cell immunosuppression.

Fig. 1. Identification of TSN as a small-molecule compound that blocks macrophage-mediated immunosuppression and activates T cells.

(A to C) Human PBMC-derived macrophages harvested from healthy donors were lentivirally transduced to express dual-gene IL-10–fLuc and CMV-rLuc gene reporters, followed by treatment with individual compounds. The cells were incubated with human glioblastoma (GBM) patient–derived tumor endothelial cells (ECs) and tumor-conditioned medium and subjected to bioluminescence analysis. (A) Experimental procedure. (B) Three rounds of screening were performed with tested compounds at different doses. Compounds in the boxes were selected for the next-round screening. FC, fold change. (C) Molecular structure of TSN. (D) Human PBMC-derived macrophages were lentivirally transduced to express IL-10–fLuc and CMV-rLuc gene reporters, treated with TSN, incubated with human tumor ECs for 2 days, and subjected to bioluminescence analysis (n = 3 human samples, mean ± SEM). Statistical analysis by one-way ANOVA. (E and F) Human macrophages were educated with or without tumor ECs in the absence or presence of TSN. Cells were analyzed by flow cytometry for expression of (E) CD206 and (F) IL-10 in macrophages. Left: Representative cell sortings with (E) all cells and (F) gated CD11b–fluorescein isothiocyanate–positive (FITC⁺) cells. Right: Quantitative results (n = 3 human samples, mean ± SEM). Statistical analysis by one-way ANOVA. (G and H) Human macrophages were treated with or without IL-4 and IL-6 in the absence or presence of TSN. Cells were analyzed by flow cytometry for expression of (G) CD206 and (H) IL-10 in macrophages. Left: Representative cell sortings with (G) CD11b-allophycocyanin (APC⁺) (H) CD11b-FITC⁺ cells. Right: Quantitative results (n = 3 human samples, mean ± SEM). Statistical analysis by one-way ANOVA. (I to N) Human tumor-educated macrophages were treated with or without 50 nM TSN. (I to K) Before and after incubation of sorted macrophages with CFSE-loaded, CD3/CD28-activated human T cells for 3 days, cells were analyzed by flow cytometry. (I) Experimental procedure. (J and K) Analysis for T cell proliferation. (J) Representative cell sorting. (K) Quantitative results (n = 3 human samples, mean ± SEM). Statistical analysis by one-way ANOVA. (L to N) After incubation of sorted macrophages with CD3/CD28-activated human T cells for 3 days, cells were analyzed by flow cytometry. Analysis for (L) CD25, (M) IFN-γ, and (N) Ki67 expression in T cells. Quantitative results (n = 3 human samples, mean ± SEM). Statistical analysis by two-tailed Student’s t test. FACS, fluorescence-activated cell sorting.

TSN reprograms immunome in tumor-educated macrophages

We next explored potential mechanism(s) by which TSN regulates Mϕ immunity. We performed transcriptome analysis by RNA sequencing (RNA-seq), and our principal components analysis (PCA) showed that TSN induced a potential plasticity switch in tumor-educated human macrophages with an altered global gene expression profile (Fig. 2A). Gene set enrichment analysis identified that the top regulated pathways in TSN-treated macrophages were related to immune systems (Fig. 2B). Specifically, TSN down-regulated the expression of M2-like alternatively polarization-associated genes, including MRC1 (CD206), CD163, and F13A1 (TGase) and of immunosuppressive cytokine genes, including IL10 and IL6 (Fig. 2C). It also up-regulated the expression of proinflammatory cytokine genes, including IL1B (IL-1β) and CXCL8 (IL-8) (Fig. 2C). In an independent study with tumor-educated human PBMC-derived macrophages treated with or without TSN, reverse transcription polymerase chain reaction (RT-PCR) analysis confirmed that TSN markedly suppressed expression of CD206, CD163, IL-10, and IL-6 and stimulated expression of IL-1β and IL-8 (Fig. 2, D and E), collectively suggesting that TSN induces a macrophage phenotype switch to acquire proinflammatory-like antitumor immunity.

Fig. 2. TSN reprograms immunome in macrophages.

Macrophages from three healthy donors were treated with tumor EC–conditioned medium in the presence or absence of 50 nM TSN. Cells were lysed, and RNAwas extracted and subjected to transcriptome analysis by RNA-seq (n = 3 humans). (A) Gene expression was subjected to PCA. (B) Gene set enrichment analysis by Reactome. Top enriched gene sets are shown. SREBF/SREBP, sterol regulatory element binding proteins; Ga-i, G alpha (i); GPCR, G protein–coupled receptor; GTPases, guanosine triphosphatases. (C) Heatmap of genes associated with macrophage polarization and immunosuppressive and proinflammatory cytokines. FPKM, fragments per kilobase of exon per million mapped fragments. (D and E) Cells were lysed, and RNA was extracted and subjected to RT-PCR analysis for (D) anti-inflammatory– and (E) proinflammatory-associated genes (n = 5 humans, mean ± SEM). Statistical analysis by two-way ANOVA test.

TSN reverses tumor immunosuppression and stimulates antitumor T cell immunity in mouse models of GBM

We next investigated the therapeutic effects of TSN on tumor immunosuppression and growth using a replication-competent avian sarcoma leukosis virus (ASLV) long terminal repeat with a splice acceptor (RCAS)/Pdgfb–driven genetically engineered Ink4a-Arf and Pten knockout mouse model that is immunocompetent and recapacitates the key features of human GBM (Fig. 3A) (27). GBM was genetically induced, followed by tumor transplantation into wildtype (WT) C57BL/6 mice. TSN blocked tumor growth during the treatment window, with a significant decrease in tumor volume at day 14 after TSN treatment (P = 0.037; Fig. 3B). This short-term, 5-consecutive-day administration of TSN led to moderately delayed tumor growth and extended animal survival (Fig. 3, C and D). Similar results were observed in an independent CT2A mouse glioma model (fig. S2). Furthermore, cytometry by time of flight (CyTOF) analysis of tumor-derived single cells showed that TSN enhanced tumor infiltration with multiple leukocytes, including CD8⁺ T cells (Fig. 3, E and F), B cells, CD4⁺ T cells, dendritic cells, and monocytes (fig. S3A). Consistent with its antitumor immune activity in vitro, TSN slightly stimulated M1-like proinflammatory macrophage polarization and significantly inhibited M2-like anti-inflammatory macrophage polarization (P = 0.0406; Fig. 3G). In accordance with these findings, flow cytometry analysis showed that TSN inhibited IL-10 expression in tumor-associated F4/80⁺ macrophages (Fig. 3H). Moreover, TSN reduced expression of immunosuppressive cytokines IL-10 and transforming growth factor–β (TGF-β) in the tumor tissues of treated mice (Fig. 3I). TSN increased the population of total CD3⁺ T cells and cytotoxic CD8⁺ T cells in the tumors (Fig. 3, J and K). TSN also enhanced the ratios of CD8⁺/CD4⁺ T cells (Fig. 3K); decreased the population of Foxp3⁺CD25⁺ regulatory T (T_reg) cells (Fig. 3L); and reduced the population of PD1⁺, Tim3⁺, and Lag3⁺ exhausted T cells in treated mice (Fig. 3M). CyTOF analysis confirmed that TSN reduced the population of Foxp3⁺CD25⁺ T_reg cells and inhibited expression of PD1, Tim3, and Lag3 in CD3⁺ and CD8⁺ T cells (fig. S3, B to D). Consistent with these findings, immunofluorescence analysis of tumor sections showed that TSN reduced CD206⁺Mac3⁺ M2-like macrophage populations and increased granzyme B (GrzB)⁺CD8⁺ active cytotoxic T cell populations in the tumors (fig. S4). In addition, we confirmed these antitumor immunity effects of TSN in a GL261 syngeneic mouse glioma model (fig. S5A). Our data showed that TSN enhanced infiltration of CD8⁺ into the tumors (fig. S5B) and reduced the populations of CD4⁺⁻CD25⁺Foxp3⁺ T_reg cells, PD1⁺ or Tim3⁺ exhausted T cells, and CD206⁺F4/80⁺ M2-like macrophages (fig. S5, D to F). These data collectively suggest that TSN reverses macrophage-mediated tumor immunosuppression and enhances infiltration of effective T cells into the tumors.

Fig. 3. TSN reverses macrophage-mediated immunosuppression and stimulates antitumor T cell immunity in vivo.

GBM was genetically induced, followed by orthotopic injection of tumor spheres into wildtype (WT) C57BL/6 mice. Mice were treated with TSN (1 mg/kg). (A) Experimental procedure. ip, intraperitoneally. (B) Tumor volumes were analyzed by bioluminescence imaging before and after TSN treatment. Left: Representative images. Right: Quantitative results (n = 6 or 7 mice per group, mean ± SEM). RU, response unit. Statistical analysis by two-way ANOVA. (C) Tumor volumes were analyzed by bioluminescence imaging. (D) Animal survival was monitored and analyzed by log-rank test. MS, median survival. (E to M) Tumors were excised and analyzed 2 days after TSN treatment. (E to G) Tumor-derived single-cell suspensions were analyzed by CyTOF. (E) Representative cell distribution by uniform manifold approximation and projection (UMAP). (F and G) Quantitative results for (F) CD8⁺ T cells and (G) M1- and M2-like macrophages. Statistical analysis by two-tailed Student’s t test (mean ± SEM, n = 5 or 6 mice per group). (H) Tumor-derived cells were analyzed by flow cytometry analysis for IL-10 expression in tumor macrophages. Left: Representative cell sorting. Right: Quantitative results (n = 5 mice per group). Statistical analysis by Mann-Whitney t test (mean ± SEM). (I) Tumor lysates were subjected to enzyme-linked immunosorbent assay analyses for IL-10 and TGF-β expression. Statistical analysis by two-tailed Mann-Whitney t test (n = 6 mice). (J to M) Tumor-derived cells were analyzed by flow cytometry for (J) CD3, (K) CD4/CD8, (L) Foxp3/CD25, and (M) Lag3, Tim3, and PD1 expression in T cells. Left: Representative cell sorting. Right: Quantitative results (n = 5 mice per group). Statistical analysis by two-tailed Mann-Whitney t test (mean ± SEM). NK, natural killer; PE, phycoerythrin; DC, dendritic cell.

TSN reprograms Mϕ immunome by inhibiting Hck and Lyn

To gain molecular insights into TSN’s mechanism of action, we identified the TSN-interacting molecules in macrophages. Considering that a series of C-28 acylated derivatives of TSN demonstrated similar in vitro biological activity to TSN (fig. S6), we covalently attached TSN to agarose beads using ketone or an OH group to identify TSN target(s). We incubated Mϕ-derived cell lysates with these TSN-conjugated beads and identified more than 1000 potential TSN-interacting proteins by mass spectrometry–based proteomic analysis after affinity chromatography (Fig. 4A). Artificial intelligence and neural network–based deep learning analysis for drug-target interaction (DTI) prediction revealed that 7 of the top 30 proteins with the highest binding scores were kinases (fig. S7), leading us to focus our studies on kinases. Further analysis showed that 63 kinases were identified as TSN-interacting proteins (Fig. 4B). Among the top 10 kinases in this list, Hck and Lyn, two Src family kinases, could be inserted by TSN into their adenosine triphosphate (ATP)–binding pockets using in silico molecular docking models (Fig. 4C). Likewise, small interfering RNA (siRNA)–mediated knockdown of Hck and Lyn inhibited Mϕ expression of arginase-1, an immunosuppressive molecule contributing to T cell inactivation (Fig. 4D), and attenuated CD206 expression in tumor-educated macrophages (Fig. 4E), supporting a role for Hck and Lyn in macrophage immunosuppression. Consistent with these findings, pharmacological inhibition of Hck and Lyn by A419259 reduced CD206 expression in tumor macrophages (fig. S8). In addition, lentivirus-mediated overexpression of Hck and Lyn partially rescued CD206 expression in TSN-treated, tumor-educated macrophages (Fig. 4F). Only overexpression of WT Hck and Lyn, but not their mutants Hck^K290M and Lyn^K275M that were unable to bind to TSN, could induce the rescue phenotypes (fig. S9). Collectively, this suggests that TSN reprograms macrophage immunity at least partially through Hck and Lyn. In accordance with these findings, surface plasmon resonance (SPR) analysis showed that TSN and its derivatives directly interacted with purified Hck and Lyn, with comparable equilibrium dissociation constants (K_d) to A419259, a well-characterized pharmacological inhibitor of pan-Src family kinases, including Hck and Lyn (Fig. 4G and fig. S6C). Moreover, TSN inhibited in vitro kinase activities of purified Hck and Lyn proteins (Fig. 4H). These data suggest that TSN interacts with and inhibits Hck and Lyn to abrogate Mϕ protumor immunity.

Fig. 4. Chemoproteomic analysis identifies Hck and Lyn as TSN’s therapeutic targets.

(A and B) TSN was covalently conjugated with agarose beads using bonded ketone or an OH group, followed by incubation with cell lysates of tumor-educated macrophages. The molecules interacting with TSN were identified by mass spectrometry after centrifugation pulldown. (A) Experimental procedure. (B) Sixty-three identified kinases’ abundances (n = 3 human samples). (C) Binding modes of Hck [Protein Data Bank (PDB) ID: 2C0I] and Lyn (PDB ID: 5XY1) superimposed to the structure of TSN (colored blue) were predicted by molecular docking simulation. (D and E) Human PBMC-derived macrophages were transfected with siRNA targeting Hck, Lyn, or control sequence and treated with GBM EC–conditioned medium. (D) Cell lysates were immunoblotted to determine expression of arginase-1, Hck, and Lyn. (E) Cells were analyzed by flow cytometry (n = 4 human samples, mean ± SEM). GAPDH, glyceraldehyde-3-phosphate dehydrogenase. (F) Human PBMC-derived macrophages were lentivirally transduced to overexpress Hck and Lyn and educated with tumor ECs in the absence or presence of 50 nM TSN. Cells were analyzed by flow cytometry for expression of CD206. Left: Representative cell sorting is shown. Right: Quantitative results (n = 3 human samples, mean ± SEM). Statistical analysis by one-way ANOVA. cDNA, complementary DNA. (G) Purified Hck and Lyn were immobilized on sensor chips, and their interaction with TSN and A419259 was analyzed by surface plasmon resonance. Left: Representative binding curves from three independent assays. Right: Calculated dissociation constant (K_d) values. (H) Purified Hck and Lyn were incubated with TSN, followed by kinase assays (n = 4 assays, mean ± SD).

TSN overcomes GBM resistance to immune checkpoint blockade

Given that TSN reverses macrophage-mediated immunosuppression and activates antitumor T cell immunity in the tumors (Fig. 3), we next sought to test combination therapy of TSN with checkpoint blockade or CAR T immunotherapy. GBM was genetically induced in mice, followed by tumor transplantation into WT C57BL/6 mice; these mice were then treated with TSN and immune checkpoint inhibitors (ICIs; anti-PD1 and anti-CTLA4 antibodies) alone or combined 10 days after inoculation (Fig. 5A). Consistent with our in vitro data (Fig. 4), TSN treatment inhibited phosphorylation of pan-Src kinases in tumor-associated macrophages (fig. S10). Our data showed that TSN or ICIs alone moderately delayed tumor growth (Fig. 5B) and extended animal survival (+2 to 3.5 days of median survival as compared with 19 days of median survival in control mice; Fig. 5C). In contrast, combination therapy delayed tumor growth (Fig. 5B) and enhanced mouse survival with >50% increased median survival (+10 days, P < 0.0001; Fig. 5C). About 10% of the combination treated mice were tumor-free survivors when the experiments reached the end point at day 50 (Fig. 5C). Flow cytometry analysis of these tumors showed that combination therapy significantly stimulated infiltration of CD8⁺ (P < 0.0001) and total CD4⁺ T cells (P = 0.0096), but not CD4⁺CD25⁺Foxp3⁺ T_reg cells (P = 0.5304), into the tumors by the end point as compared with control mice (Fig. 5, D and E). Moreover, combination therapy enhanced the populations of active GrzB⁺ T cells, decreased M2-like CD206⁺F4/80⁺ macrophages in the tumors (Fig. 5, F to H), and reduced the area of hemorrhagic necrosis as compared with control mice (fig. S11A), a hallmark feature of GBM. In a parallel study, similar and apparently more robust therapeutic results were observed in an independent GL261 syngeneic mouse GBM model as compared with the genetically induced model (fig. S11, B to D). Combination treatment of TSN with ICIs resulted in complete therapeutic responses in more than half of the treated mice.

Fig. 5. Combination therapy with TSN and immune checkpoint inhibitors inhibits GBM growth and improves animal survival.

GBM was genetically induced, followed by transplantation of tumor spheres into WT C57BL/6 mice. Mice were treated with TSN and immune checkpoint inhibitors (ICIs) 10 days after inoculation for five consecutive days. (A) Experimental procedure. (B) Tumor volumes were analyzed by bioluminescence imaging. (C) Animal survival was monitored and analyzed by log-rank test (n = 16 to 19 mice per group, pooled from two experiments). (D to H) Tumor-derived single cells were analyzed by flow cytometry for (D) CD8⁺ and CD4⁺ cells of CD45⁺CD3⁺ T cells, (E) CD4⁺CD25⁺Foxp3⁺ T_reg cells of CD45⁺CD3⁺ T cells, (F) GzmB⁺, Tim3⁺, and PD1⁺ cells of CD45⁺CD3⁺CD8⁺ T cells, (G) CD11b⁺F4/80⁺ macrophages of CD45⁺ hematopoietic cells, and (H) CD206⁺ and PDL1⁺ cells of CD11b⁺F4/80⁺ macrophages. Statistical analysis by two-tailed Mann-Whitney t test (mean ± SEM, n = 6 or 7 mice per group). (D) Left: Representative cell sorting. Right: Quantitative results.

In addition, we tested the effects of TSN on systemic immunosuppression in vivo in mice bearing genetically induced GBM tumors. Our data showed that tumors suppressed the cell proliferative capacity of T cells derived from peripheral blood and thymus, but not spleen and bone marrow, in the end stage of cancer when mice develop severe neurological symptoms, including domehead, hemiparesis, or loss of more than 20% of body weight (fig. S12, A and B). Tumors also enhanced the populations of circulating CD4⁺CD25⁺Foxp3⁺ T_reg cells and CD11b⁺Gr1⁺ myeloid-derived suppressor cells (MDSCs) (fig. S12, C and D). TSN did not reverse these inhibitory effects in circulating and thymus T cells and MDSCs, as indicated by lack of increase in cell proliferative capacity of peripheral T cells and by lack of decrease in circulating MDSC numbers in TSN-treated mice compared with control mice (fig. S12, A–D), suggesting selective activity of TSN on intratumoral immunity but not on systemic immunosuppressive phenotypes.

TSN sensitizes GBM tumors to CAR T immunotherapy in mice

We lastly investigated the effects of TSN on T cell therapy with a fully murine CAR system in a setting that recapitulates human GBM. GBM was genetically induced, and tumor spheres were transduced to express mouse Egfrviii before transplantation into recipient WT mice, leading to Egfrviii expression in about 30 to 40% of the tumor cells (Fig. 6A), which is comparable to human patients with GBM (28). Our data with adoptive rLuc⁺ and 5-chloromethylfluorescein diacetate (CMFDA) labeled CAR T cells showed that TSN promoted CAR T cell infiltration (Fig. 6, B and C) and inhibited CAR T cell exhaustion, as indicated by reduced expression of Lag3, PD1, and Tim3 in tumor-associated CAR T cells (Fig. 6D), likely due to TSN-induced reversal of tumor immunosuppression (Fig. 3). Moreover, CAR T immunotherapy alone did not substantially affect tumor growth and mouse survival (Fig. 6, E and F), consistent with limited basal cell infiltration after T cell infusion (Fig. 6B). In contrast, TSN treatment sensitized tumors to CAR T immunotherapy, as indicated by a delayed tumor growth in 40% of the mice (Fig. 6E). Likewise, the combination therapy significantly enhanced mouse survival (P < 0.0001; Fig. 6F). Thirty percent of the mice in the combination therapy group survived by 36 days after tumor induction, when all of the mice in the other groups had died. In addition, we confirmed the therapeutic effects of the combination therapy in the GL261 model (fig. S13). Furthermore, triple combination of TSN, ICIs, and Egfrviii CAR T cell therapy induced significant therapeutic efficacy (P < 0.0001, compared with control CAR T treatment group), resulting in extended animal survival and complete therapeutic responses in two of three of the treated GL261-bearing mice (fig. S13). Together, these results indicate that TSN treatment reduces GBM resistance to CAR T immunotherapy.

Fig. 6. TSN sensitizes GBM tumorsto CAR T cell immunotherapy.

GBM was induced in immunocompetent mice by transplantation of Egfrviii⁺fLuc⁺ glioma tumor cells into WT C57BL/6 mice, followed by TSN treatment and adoptive rLuc⁺ CAR T cell transfer at days 8 and 14, respectively. (A) Experimental procedure. iv, intravenous. (B) Two days after CAR T cell therapy, T cell trafficking was analyzed by bioluminescence imaging. Left: Representative images. Right: Quantitative results for tumor volumes and T cell engraftment (mean ± SEM, n = 6 to 8 mice per group). Statistical analysis by two-tailed Mann-Whitney t test. (C and D) Two days after infusion with CMFDA-labeled CAR T cells, tumor-derived single-cell suspensions were analyzed by flow cytometry for (C) infiltration and (D) exhaustion of CAR T cells. Left: Representative cell sorting. Right: Quantified results (means ± SEM, n = 6 mice per group). Statistical analysis by two-tailed unpaired Student’s t test. (E) Tumor volumes were analyzed by bioluminescence imaging. (F) Animal survival was monitored and analyzed by log-rank test (n = 8 to 10 mice per group).

DISCUSSION

Reversal of intratumoral immune suppression is a much-needed therapeutic strategy to guarantee success of T cell–based cancer immunotherapy. However, no effective approaches are currently available in the clinic, particularly for targeting tumor Mϕs. Treatments manipulating Mϕ proliferation, differentiation, and activation pathways that are mediated through macrophage colony-stimulating factor (CSF-1), PI 3-kinase, Toll-like receptor 4, CD40, and CD47 have recently been exploited, showing small benefits or are still under evaluation (18–21). Here, we identify that the small-molecule TSN exhibits efficacy against Mϕ-mediated immunosuppression at relatively low doses, such as 50 nM in vitro and 1 mg/kg body weight in vivo, with minimal effects on cell viability in normal Mϕs. These data suggest that TSN may serve as a tumor Mϕ-selective targeting therapeutic compound for improving cancer immunotherapy.

Our work suggests that TSN activates antitumor immunity by an Hck- and Lyn-dependent genetic mechanism that reprograms global expression of immune-associated genes in macrophages. Previous studies have established a critical role for Src family kinases in macrophage functions including adhesion, migration, phagocytosis, proliferation, activation, and cytokine production (29–32). More recent work suggests that Hck and Lyn regulate macrophage-mediated inflammation progression and resolution (33, 34). Our study shows that Hck and Lyn are critical for immunosuppressive polarization in tumor macrophages, consistent with a protumor role of Hck for myeloid cell–mediated tumor progression and immunosuppression (35, 36). In addition, either pharmacological inhibition by A419259 or genetic knockdown of Hck and Lyn partially reduced CD206 expression in tumor macrophages, showing less potency than TSN (Figs. 1E and 4E, and fig. S8), which suggests the potential existence of additional mechanisms underlying TSN’s therapeutic effects.

TSN is a triterpenoid originally extracted from medicinal herb Melia toosendan Sieb that was traditionally used as a parasiticide and insecticide in Eastern medicine (37). Recent work suggests that TSN at higher doses has anticancer activity that is related to apoptosis induction and cell cycle arrest in tumor cells (38–42). We identify that TSN activates antitumor immunity by altering the global expression of immune-associated genes and reversing immunosuppression in macrophages. This macrophage-reprogramming activity might also contribute to the observed compound’s antiparasite functions by stimulating proinflammatory gene expression in phagocytes.

Regarding the limitations of this study, therapeutic exploitation of combined TSN and immunotherapy was conducted in murine GBM models that do not capture the diverse heterogeneity of human glioma. Future work using immune-humanized patient-derived xenograft models will be needed to validate our results and to provide further information for clinical treatment in humans. In addition, it is unclear whether TSN also modulates immune activity of other myeloid cells, including dendritic cells and MDSCs, to exert its potential proinflammatory effects on tumor immunity. Chemical modification and optimization that improve its delivery across the blood-brain barrier and pharmacokinetics and pharmacodynamics profiles may further enhance TSN’s therapeutic efficacy as a partner drug for use in cancer immunotherapy combinations. Hence, our efforts to optimize druggability of TSN have led to the production of several C-28 acylated derivatives that show comparable in vitro activity (fig. S6), suggesting C-28 as a potentially feasible modification site for further drug development.

In summary, our work establishes TSN as a small-molecule compound that reverses macrohage-mediated tumor immunosuppression and activates antitumor T cell immunity by targeting Hck and Lyn. The efficacy in GBM mouse models warrants the further development of TSN and its derivatives. TSN treatment may provide a key opportunity to overcome solid tumor resistance to T cell–based immunotherapy and warrants further investigation.

MATERIALS AND METHODS

Study design

The overall objectives were to determine the therapeutic effects of TSN on tumor immunotherapy resistance and to define the drug mechanism of action in mouse models of GBM. We aimed to screen a chemical library and identify TSN as a compound that reverses immunosuppressive phenotypes in Mϕs in vitro. We analyzed gene expression signature in TSN-treated tumor ECs and investigated the effects of TSN on global expression of immune-associated genes by RNA-seq. We tested the in vivo therapeutic effects of TSN on tumor growth and immune compositions in multiple syngeneic mouse GBM models, analyzed by CyTOF and flow cytometry. We further identified the underlying mechanisms by mass spectrometry–based chemoproteomic analysis of pulldown proteins with TSN-conjugated beads and investigated the TSN interaction with Hck/Lyn and regulation of their activity by SPR, mutagenesis, kinase activity assay, and cell-based approaches. Last, we interrogated the efficacy of combined therapy with TSN plus ICIs or CAR T cells in mouse GBM models. The in vitro experiments were carried out with at least three replicates. All experiments were randomized and blinded where possible. Sample sizes were determined on the basis of expected effect sizes from pilot experiments and power analysis. In general, group sizes of three or more human samples were used, and mice of similar age, weight, and sex were grouped randomly with the number per group of 4 to 19.

Human monocyte isolation and treatment

PBMC-derived monocytes were provided from healthy human volunteer donors from ages 18 to 64 of all genders, races, and ethnicities by Human Immunology Core at the University of Pennsylvania. Blood samples were obtained for research purposes with informed consent by each donor, and the respective regulations were followed under a University Institutional Review Board–approved protocol. Monocytes were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) and treated with CSF-1 (10 ng/ml; BioLegend, 574806) to induce macrophage differentiation. Cells were cocultured with human GBM patient–derived tumor ECs that were isolated and verified as previously described (27, 43) or treated with tumor EC–conditioned medium or IL-4 (100 ng/ml; BioLegend, 574004) or IL-6 (BioLegend, 570806) in the presence of TSN (Carbosynth, FT65803) and its derivatives (provided by M. Parker, Fox Chase Chemical Diversity Center) or trihydrochloride (A419259, Cayman Chemical, 29506).

Human EC isolation and culture

Surgical specimens from human patients with GBM from ages 48 to 83 of all genders, races, and ethnicities were collected at the Department of Neurosurgery of the Hospital of the University of Pennsylvania. The collection of human tissues in compliance with the tissue banking protocol was approved by the University of Pennsylvania Institutional Review Board, and written informed consent was obtained from each participant. ECs were isolated and verified as previously described (27, 43). Tumor-derived single-cell suspensions were prepared by the tissue bank. Cell suspensions were subjected to magnetic-activated cell sorting (MACS) with anti-CD31 antibody–conjugated magnetic beads (Miltenyi Biotech, 130-091-935). Sorted ECs were verified by Dil-Ac-LDL (Alfa Aesar, J65597) absorption and von Willebrand factor (vWF) staining. Cells were checked and showed no mycoplasma contamination. All cells were used between passages 2 and 6.

Drug screening assay

Human IL-10 promoter region was amplified by PCR with primer sequences 5′-AATTGGTACCTCAACTTCTTCCACCCCATC-3′ and 5′-AATTGCTAGCGCCTTCTTTTGCAAGTCTGT-3′, followed by subcloning into pGL3-fLuc vector at Kpn I and Nhe I sites. Plasmid was verified by DNA sequencing. Human embryonic kidney (HEK) 293T cells were transfected with envelope-encoding and IL-10–fLuc or CMV-rLuc plasmids using Lipofectamine 3000 (Thermo Fisher Scientific, L300001) in Opti-minimal essential medium (Thermo Fisher Scientific, 31985–070). Human monocytes were transduced with the lentivirus-containing medium, treated with chemical compounds, and cocultured with tumor ECs in tumor-conditioned medium (collected from U251 human glioma cells cultured under hypoxia with 1% O₂). Bioluminescence was measured using a Synergy H4 Hybrid plate reader (BioTek).

CFSE assay

Human T cells were obtained from healthy human volunteer donors by Human Immunology Core at the University of Pennsylvania. The collection protocol was approved by the University of Pennsylvania Institutional Review Board, and written informed consent was obtained from each participant. T cells were incubated with CellTrace CFSE solution (Thermo Fisher Scientific, C34554) for 20 min at 37°C. After treatment, cells were analyzed using a Canto II flow cytometer (BD Biosciences).

Cell proliferation assay

Human PBMC-derived monocytes and T cells, astrocytes (Lonza, CC-2565), and U251 and U87 glioma cells (Sigma-Aldrich, 09063001 and 89081402) were seeded on 96-well plates (3000 cells per well) and treated with TSN at different doses for 24 hours. Cell proliferation was measured using a CyQUANT Direct Cell Proliferation Assay kit (Thermo Fisher Scientific, C35011) according to the manufacturer’s instruction. Bioluminescence was detected using a Synergy H4 Hybrid luminescent plate reader (BioTek).

RNA-seq analysis

Human monocytes were pretreated with CSF-1 (10 ng/ml) for 3 days and treated with tumor EC–conditioned medium in the presence or absence of 50 nM TSN for 24 hours. RNA was extracted using TRIzol (Thermo Fisher Scientific, 15596026) and purified using an RNeasy kit (Qiagen, 74004) according to the manufacturers’ instructions. The library was constructed with a TruSeq mRNA Stranded Kit (Illumina) followed by quality analysis with DNA Nano assay chips using a 2100 bioanalyzer (Agilent) and was subjected to next-generation sequencing (HiSeq 2500) at Next-Generation Sequencing Core of University of Pennsylvania. The sequences were aligned to the UCSC hg38 reference genome using RNA-Star (version 2.4.2a; https://github.com/alexdobin/STAR).The gene expression was normalized and calculated as fragments per kilobase million mapped reads values by Cufflinks (version 2.2.1; http://cole-trapnell-lab.github.io/cufflinks/releases/v2.2.1/) with Gencode version 22 gene annotations (www.gencodegenes.org/human/release_22.html).

Real-time RT-PCR analysis

RNA was extracted using TRIzol (Thermo Fisher Scientific, 15596026) according to the manufacturer’s instructions. Real-time RT-PCR was conducted using SuperScript III First-Strand Synthesis SuperMix (Life Technologies) and Fast SYBR Green Master Mix (Applied Biosystems). The primers used were as follows: MRC1, 5′-CGTTTACCAAATGGCTTCGT-3′ (forward) and 5′-GTACCCATCCTTGCCTTTCA-3′ (reverse); CD163, 5′-GCATAGTAACTGTACTCACCAACA-3′ (forward) and 5′-CCA-GAACACATATTCCCTCCAC-3′ (reverse); IL10, 5′-GCCTAA-CATGCTTCGAGATC-3′ (forward) and 5′-TGATGTCTGGGTCTTGGTTC-3′ (reverse); IL6, 5′-GCAGAT-GAGTACAAAAGTCCTGA-3′ (forward) and 5′-TTCTGTGCCTGCAGCTTC-3′ (reverse); CXCL8, 5′-GAGACAGCAGAGCACACAAG-3′ (forward) and 5′-CTTCACACAGAGCTGCAGAA-3′ (reverse); IL1B, 5′CAGCCAATCTTCATTGCTCAAG-3′ (forward) and 5′-GAA-CAAGTCATCCTCATTGCC-3′ (reverse).

Mass CyTOF

Tumor-derived single-cell suspensions were prepared by mechanical dissociation with a gentleMACS Dissociator (Miltenyi Biotech) and enzymatic digestion with collagenase II and dispase II. Cells were incubated with heavy metal–conjugated antibodies (Fluidigm) for 30 min and treated with 25 μM cisplatin for 3 min at room temperature. Cells were washed with wash buffer, fixed with 1.6% paraformaldehyde, and stained with Cell-ID Intercalator-Ir (Fluidigm), followed by analysis with a CyTOF2 mass cytometer (Fluidigm). Data were analyzed and visualized by R (version 4.1.2).

T cell isolation, culture, and treatment

Mouse T cells

T cells were isolated from spleens of C57/B6 mice by mechanical dissociation with a gentleMACS Dissociator (Miltenyi Biotech). T cells were cultured in RPMI 1640 medium supplemented with 10% FBS. For retrovirus transduction, cells were treated with CD3/CD28 antibodies (5 μg/ml; BioLegend, 100302 and 102102) and recombinant IL-2 (100 IU/ml; Corning, 354043) for 2 days. A retrovirus expressing rLuc-tdTomato, mouse Egfrviii-CAR T cells, or control plasmid mouse stem cell virus–based splice-gag (pMSVG) empty vector sequence was prepared using Phoenix cells transfected with a pCL-Eco helper plasmid and Lipofectamine 2000 transfection reagent (Life Technologies; 11668–019). T cells were retrovirally transduced as described previously (28). Cells were immunostained with a goat anti-human F(ab′) 2-biotinylated antibody (Jackson ImmunoResearch, 109-065-006) to detect CAR expression using a FACSCanto II flow cytometer (BD Biosciences), and the data were analyzed using FlowJo software (28).

Human T cells

PBMC-derived T cells were provided from healthy human volunteer donors from ages 18 to 64 of all genders, races, and ethnicities by Human Immunology Core at the University of Pennsylvania. Blood samples were obtained for research purpose with informed consent by each donor, and the respective regulations were followed under a University Institutional Review Board–approved protocol. T cells were cultured in RPMI 1640 medium supplemented with 10% FBS. For T cell activation, cells were cultured with a CD3/CD28 T cell activator (StemCell, 10971) and recombinant IL-2 (BD Biosciences, 354043) for 2 days.

Mouse tumor models and treatment

Genetically engineered GBM was induced in mice as described previously (27, 44–46). Briefly, chicken DF-1 fibroblasts [American Type Culture Collection (ATCC), CRL-12203] were transfected with replication-competent ASLV long terminal repeat with a splice acceptor (RCAS)–Pdgfb and RCAS-Cre plasmids to produce retrovirus, which was orthotopically injected into Ntv-a;Ink4a-Arf^−/−;Pten^fl/fl;LSL-luc mice (2 months old, half male and half female, provided by E. Holland, Fred Hutchinson Cancer Research Center) to induce GBM through RCAS/n-tva–mediated gene transfer. Tumors were isolated and subjected to mechanical dissociation with a gentleMACS Dissociator (Miltenyi Biotech) and enzymatic digestion with collagenase II and dispase II to obtain single-cell suspensions. For retroviral transduction, the medium supernatant was collected from DF-1 cells that were transfected to express RCAS, and mouse Egfrviii was collected, followed by filtration with a 0.45-μm sterilized filter and concentration using a Retro-X Concentrator reagent (Takara, 631456). After centrifugation, the virus pellet was suspended in mouse stem cell medium and incubated with tumor-derived sphere cells. About 8-week-old WT C57BL/6J mice (half male and half female; Jackson Laboratory, 000664) were orthotopically and stereotactically injected with 10⁵ tumor cells. For the GL261 syngeneic model, mouse GL261 glioma cells (PerkinElmer, 134246) were lentivirally transduced to express fLuc and/or mouse Egfrviii and green fluorescent protein (GFP), and GFP⁺ cells were harvested by cell sorting. RCAS, GL261, or CT-2A-luc (2 × 10⁵; Sigma-Aldrich, SCC195) cells in a total volume of 2 μl were orthotopically injected into C57BL/6J mice. Tumor-bearing mice were intraperitoneally treated with TSN (1 mg/kg), anti-PD1 antibody (100 μg per mouse; BioXCell, BE0146), anti-CTLA4 antibody (100 μg per mouse; BioXCell, BE0131), or control rat immunoglobulin G (IgG; BioXCell, BE0090) in saline or administrated with CAR T cells (5 × 10⁶ cells per mouse) through tail vein. For T cell imaging, mice were injected with mouse T cells expressing tdTomato-rLuc (3 × 10⁶ cells per mouse) through tail vein 14 days after GBM induction. Mice were imaged after retro-orbital injection of coelenterazine (4 mg/kg; Promega), followed by secondary imaging after retro-orbital injection of D-luciferin (150 mg/kg; GoldBio). Postinjection survival was monitored for up to 50 days. Tumor growth was monitored by whole-body bioluminescence using an IVIS 200 Spectrum Imaging System after retro-orbital injection of luciferin (150 mg/kg; GoldBio). Mice were euthanized when exhibiting severe GBM symptoms, including domehead, hemiparesis, or more than 20% of body weight loss. Mice were randomized to receive treatment, and the investigators were not blinded. All animals were housed in the Association for the Assessment and Accreditation of Laboratory Animal Care–accredited animal facility of the University of Pennsylvania. All experiments with mice were performed in accordance with a protocol approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania.

Cell isolation for systemic immunosuppression assay

Blood, spleens, thymi, and bone marrow were collected from GBM-bearing mice exhibiting severe GBM symptoms. Blood was collected from abdominal aortas and mixed with 1 ml of phosphate-buffered saline (PBS) containing 5 μM EDTA. Single-cell suspensions of spleen and thymus were prepared by smashing tissues through a 40-μm cell strainer. Single-cell suspensions of bone marrow were prepared by flushing the femur and tibia through a 40-μm cell strainer. Red cells were removed with ammonium-chloride-potassium lysis buffer (Thermo Fisher Scientific, A1049201). Cells were cultured in RPMI 1640 medium (Life Technologies) with 10% FBS and treated with recombinant IL-2 (100 IU/ml; Corning, 354043) and CD3/CD28 antibodies (5 μg/ml; BioLegend, 100302 and 102102) for 3 days to activate T cells.

Flow cytometry

Human monocytes or mouse tumor–derived single-cell suspensions were immunostained with fluorescence dye–conjugated antibodies against CD206 (1:100; BD Biosciences, 551135), IL-10 (1:50; BD Biosciences, 562036), IFN-γ (1:50; BioLegend, 502505 or 505808), Ki67 (1:100; Thermo Fisher Scientific, 69-5698-80), CD25 (1:100; BioLegend, 102016, 356107, or 102025), CD3 (1:100; BioLegend, 100233), CD4 (1:100; BioLegend, 100540), CD11b (1:200; BioLegend, 101205 or 101212), CD11b (1:200; eBioscience, 69-0112-80), CD8a (1:100; BioLegend, 100706, 100707, or 301031), CD45 (1:200; eBioscience, 48-0451-82 or 25-0453-82), F4/80 (1:200; BioLegend, 123114), CD69 (1:100; BioLegend, 104511), IL-10 (1:50; BioLegend, 505009), Foxp3 (1:50; BioLegend, 126405 or eBioscience, 12-5773-80), LAG3 (1:100; eBioscience, 11-2231-80 or BioLegend, 125225 or 125209), PD1 (1:100; BioLegend, 135207 or 135223), TIM3 (1:100; eBioscience, 48-5871-82 or BioLegend, 134003), CD206 (1:100; BioLegend, 141707), PDL1 (1:100; BioLegend, 124307 or 124333), Ly6C (1:100; BioLegend, 128005), Gr-1 (1:200; BioLegend, 108423), MHCII (1:200; eBioscience, 47-5321-80), GrzB (1:50; BioLegend, 372203), or control IgG. For intracellular staining, cells were incubated with GolgiStop solution (1:100; BD Biosciences, BDB554715) for 4 hours at 37°C. For analysis of tumor-associated CAR T cells, CAR-transduced mouse T cells were incubated with CellTracker Green CMFDA Dye (Thermo Fisher Scientific, C34554) for 20 min at 37°C, followed by administration (5 × 10⁶ cells per mouse) through the tail vein in GBM-bearing mice treated with or without TSN (1 mg/kg for 5 days). After 2 days, tumors were excised and single-cell suspensions were stained with fluorescence dye–conjugated antibodies against CD45 (1:200; eBioscience, 48-0451-82), Lag3 (1:100; BioLegend, 125225), PD1 (1:100; BioLegend, 135223), and Tim3 (1:100; BioLegend, 134003). Cells were analyzed using a FACSCanto II flow cytometer (BD Biosciences) and FlowJo software (version 10.4).

siRNA treatment

Human monocytes were transfected with siRNAs targeting Hck, Lyn, or control siRNA (Qiagen, 1027280) using an Amaxa 4D Nucleofector (Lonza) with program EA-100. The sequences of used siRNA are as follows: Hck #1 (CD.Ri.376216.13.1), 5′-GGA-GAUACCGUGAAACAUUACAAGA-3′ and 3′-UCCCU-CUAUGGCACUUUGUAAUGUUCU-5′; Hck #2 (CD.Ri.376216.13.2), 5′-GAUACCGUGAAACAUUACAAGAUCC-3′ and 3′-CUCUAUGGCACUUUGUAAUGUUCUAGG-5′; Lyn #1 (CD.Ri.376225.13.1), 5′-GCAAUCAACUUUGGAUGUUUCAC-TA-3′ and 3′-UUCGUUAGUUGAAACCUACAAAGUGAU-5′; Lyn #2 (CD.Ri.376225.13.3), 5′-AGAUUCUAAUCUCUGAA-GAACCUTA-3′ and 3′-AUUCUAAGAUUAGAGACUUCUUG-GAAU-5′.

Lentiviral transduction

Human WT or mutated Hck and Lyn complementary DNA were subcloned into pLV expression vectors (VectorBuilder). HEK293T cells (ATCC) were transfected with lentivirus packing vectors and expression vectors. The virus-containing supernatant was collected and centrifuged to remove the cell debris and filtered with membranes with a pore size of 0.45 μm. Treated human monocytes were transduced with lentivirus with polybrene (8 μg/ml; Millipore) for 2 days.

Chemical proteomics

TSN was immobilized to argarose beads using SepSphere small-molecule alcohol or ketone immobilization kits (CellMosaic, CM71007 and CM71010) following the manufacturer’s instructions. Cell lysates from human monocytes were incubated with control or TSN-conjugated agarose beads for 1 hour at 4°C. Beads were centrifuged and washed with lysis buffer three times and digested with trypsin at 37°C overnight. The peptides were desalted using C18 micro spin columns and dried using a vacuum centrifuge before reversed-phase liquid chromatography–tandem mass spectrometry analysis. Samples were resolved using an EASY-nLC System (Thermo Fisher Scientific). Peptides were eluted using a linear gradient method, and data were acquired using an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific). A full-scan mode ranging 350 to 2000 mass/charge ratio was conducted using MaxQuant software and annotated by UniProt database (www.uniprot.org/) at the Proteomics Core Facility at Children’s Hospital of Philadelphia.

Machine learning for DTI prediction

A deep learning model for DTI was performed using DeepPurpose (https://github.com/kexinhuang12345/DeepPurpose). Briefly, an encoder-decoder deep learning model was trained using the K_d value from BindingDB database, and the “Transformer” and “CNN” algorithms were used for drug or target encoding, respectively. The whole data were randomly split with 70% as training, 20% as testing, and 10% as validation dataset. The target proteins obtained from chemical proteomic analysis were translated into amino acid sequences, and TSN small molecule was converted into SMILES fingerprints. Binding affinity between the target proteins and TSN was predicted using these inputs and the pretrained model.

Molecular docking

The molecular docking studies were performed using Discovery Studio 3.5 (BIOVIA Foundation) between TSN and HCK (2C0I) or LYN (5XY1). The structure of macromolecules was prepared by removing the ligands, cofactors, and solvent molecules and adding polar hydrogen. The coordinates of active sites or original ligands were used for docking analysis. The TSN Protein Data Bank (PDB) file was downloaded from PubChem (https://pubchem.ncbi.nlm.nih.gov/). The prepared receptors and TSN were docked using libdock with best conformation, and all other parameters were set as default. Different poses with best binding affinity were visualized by the BIOVIA Discovery Studio Visualizer.

Immunoblot

For cell sample preparation, cells were lysed with an NP-40 buffer with protease inhibitor cocktail (Roche, 11697498001). For tissue sample preparation, single-cell suspensions derived from mouse tumors were incubated with anti-CD11b antibody–conjugated microbeads (1:100; Miltenyi Biotech, 130-049-601) for 15 min at 4°C and separated by MACS column with a separator. The eluted cells were lysed with NP-40 buffer with protease inhibitor cocktail and phosphatase inhibitor (Thermo Fisher Scientific, 78428). Total protein (20 μg) was resolved by 4 to 15% precast SDS–polyacrylamide gel electrophoresis (Bio-Rad) and followed by transfer. Polyvinylidene difluoride membranes were blotted with anti–phosphoSrc (Tyr⁴¹⁶; Cell Signaling Technology, 6943), anti-Hck (1:1000; Cell Signaling Technology, 14643; ABclonal, A14537), anti-Lyn (1:1000; Cell Signaling Technology, 2796), anti-FLAG (1:1000; GenScript, A00187-100), anti–arginase 1 (1:500; Santa Cruz Biotechnology, sc-20150), or anti–glyceraldehyde-3-phosphate dehydrogenase (1:3000; Cell Signaling Technology, 5174) antibody overnight at 4°C. Proteins were detected with horseradish peroxidase–conjugated secondary antibodies (Bio-Rad) and enhanced by enhanced chemiluminescence development (GE Healthcare, RPN2232).

Immunofluorescence

After deparaffinization and rehydration, mouse tumor sections were incubated with antigen retrieval solution (Dako, S1699) at 95°C for 20 min. Sections were blocked with 5% horse serum for 1 hour and incubated with anti-Mac3 (1:10; BD Biosciences, 550292), anti-CD206 (1:100; R&D Systems, AF2535), anti-CD8a (1:100; Cell Signaling Technology, 98941), or anti-GrzB (1:100; Abcam, ab255598) antibody overnight at 4°C. After washing with PBS, sections were stained with Alexa Fluor 488– or Alexa Fluor 568–conjugated IgGs (1:500; Life Technologies) or tyramide signal amplification plus systems (Akoya Biosciences, NEL744001KT/NEL741001KT) for 1 hour at room temperature. Images were acquired using an Axio Imager microscope (Zeiss) equipped with an AxioCam 506 monochrome charge-coupled device camera (Zeiss).

In vitro kinase assay

Kinase assay was performed using an ADP-Glo system (Promega, V9101) according to the manufacturer’s instructions. Briefly, recombinant human Hck (Promega, VA7186) and Lyn (Promega, VA7486) were incubated with 200 μM ATP, substrates, and TSN (dose ranging from 0 to 10 μM) for 1 hour at room temperature. After incubation, ADP-Glo reagent was added to stop the reaction, and luminescence was measured by adding detection reagent using a plate reader (Synergy H4 Hybrid, BioTek).

SPR assay

Purified Lyn (Promega, VA7486) and Hck (Promega, VA7186) and His-tagged Lyn (EuProtein, EP8740562) were immobilized on a carboxymethyldextran sensor chip (CMD700M, GE Healthcare) through amine coupling. Briefly, the chip was activated with 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC)/N-hydroxy-succinimide (NHS) buffer, and proteins were immobilized at pH at least one log unit below their isoelectric points, followed by chip blocking of the remaining activated sites with ethanolamine. Kinase binding to TSN and its derivatives and A419259 (Cayman Chemical, 25605) was measured by SPR analysis using a Biacore T200 SPR instrument (GE Healthcare) as described previously (47).

Statistical analysis

Raw, individual-level data are presented in data file S1. Comparisons between two groups were performed using unpaired two-tailed Student’s or Mann-Whitney t test. Differences among more than two groups were assessed using one- or two-way analysis of variance (ANOVA), followed by post hoc corrections for multiple comparisons (Tukey’s or Šidák’s). Kaplan-Meier survival curves were generated using Prism software (GraphPad, version 9.0), and log-rank test was performed to assess statistical significance between groups in mouse experiments. A two-sided P value lower than 0.05 was considered significant. n indicates the number of biological replicates, all bars within the graphs represent mean values, and the error bars represent SDs or SEMs.

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. The RNA-seq data generated and analyzed in this manuscript have been deposited in the Gene Expression Omnibus under accession number GSE196452. The proteomic data in this study have been deposited in the Proteomics Identifications Database (PRIDE) under accession number PXD039067.