Abstract

   Genomic alteration can reshape tumor microenvironment to drive tumor
   malignancy. However, how PTEN deficiency influences
   microenvironment-mediated cell-cell interactions in glioblastoma (GBM)
   remains unclear. Here, we show that PTEN deficiency induces a symbiotic
   glioma-M2 macrophage interaction to support glioma progression.
   Mechanistically, PTEN-deficient GBM cells secrete high levels of
   galectin-9 (Gal-9) via the AKT-GSK3β-IRF1 pathway. The secreted Gal-9
   drives macrophage M2 polarization by activating its receptor Tim-3 and
   downstream pathways in macrophages. These macrophages, in turn, secrete
   VEGFA to stimulate angiogenesis and support glioma growth. Furthermore,
   enhanced Gal-9/Tim-3 expression predicts poor outcome in glioma
   patients. In GBM models, blockade of Gal-9/Tim-3 signaling inhibits
   macrophage M2 polarization and suppresses tumor growth. Moreover,
   α-lactose attenuates glioma angiogenesis by down-regulating
   macrophage-derived VEGFA, providing a novel antivascularization
   strategy. Therefore, our study suggests that blockade of Gal-9/Tim-3
   signaling is effective to impair glioma progression by inhibiting
   macrophage M2 polarization, specifically for PTEN-null GBM.
     __________________________________________________________________

   PTEN-deficient glioma promotes macrophage M2 polarization via
   Gal-9/Tim-3 signaling, serving as a viable target in glioblastoma.

INTRODUCTION

   Genomic profiling studies have indicated that glioblastoma (GBM) can be
   classified into four subtypes, including the mesenchymal, classical,
   neural, and proneural subtypes ([50]1). Among the four subtypes,
   mesenchymal GBM exhibits the most abundant stromal components ([51]2),
   most of which are tumor-associated macrophages (TAMs) and they
   generally appear to be type 2 polarized macrophages (M2) ([52]3). TAM
   populations are bone marrow–derived infiltrating myeloid cells ([53]4)
   and constitute as much as 30 to 50% of all cells in human GBM tumors
   ([54]5). This cell population plays a key role in glioma progression.
   Mesenchymal GBM is a high-grade malignant tumor that typically harbors
   gene mutations in phosphate and tension homology deleted on chromosome
   ten (PTEN), TP53, and NF1 ([55]1). PTEN deletion or mutation results in
   enhanced recruitment of macrophages into the tumor microenvironment
   (TME) ([56]6), suggesting that macrophage-targeted therapy may
   specifically work for PTEN-null glioma.

   The interplay between cancer cells and macrophages plays a crucial role
   in tumor development, with functions in supporting angiogenesis,
   nurturing cancer stem cells, and promoting an immunosuppressive TME
   ([57]7, [58]8). The TME is shaped by tumor cell–intrinsic signaling
   pathways and secreted factors ([59]9, [60]10). Some studies reported
   that CD163^+ macrophages are essential for promoting angiogenesis by
   expressing high levels of vascular endothelial growth factor A (VEGFA)
   ([61]11), and VEGFA derived from tumor-infiltrating myeloid cells is
   important for initiating vascularization in gliomas ([62]12). Hence,
   macrophage-targeted strategies might be a novel therapeutic
   antivascularization approach. Now, immunotherapy has not been widely
   applied to glioma because of the unique immune microenvironment
   characterized by a paucity of T cells but abundant M2-like macrophages
   ([63]13). Many approaches have been attempted to eliminate TAMs or
   reprogram macrophages to perform antitumor functions ([64]14). One
   strategy used colony-stimulating factor 1 receptor (CSF1R) inhibition
   to block glioma progression by reducing macrophage M2 polarization
   ([65]15), but acquired resistance was observed in animal model
   experiments ([66]16) and limited efficacy was found in clinical trials
   ([67]17). Thus, identifying other strategies to improve the
   effectiveness of macrophage-targeted therapies for glioma patients is
   critical. Given that PTEN-null glioma can recruit a large number of
   macrophages, clarifying how PTEN deficiency in glioma cells modulates
   macrophage M2 polarization might provide new targets for precise
   immunotherapy in specific GBM subtypes or genotypes.

   In our study, we explored whether and how PTEN alterations in GBM cells
   influence macrophage polarization, and how the infiltrating macrophages
   in turn function to support the growth and angiogenesis of glioma. Our
   aim is to identify practical therapeutic targets functioning at the
   maintenance of the symbiotic glioma-macrophage interactions.

RESULTS

PTEN deficiency facilitates macrophage infiltration and macrophage M2
polarization in GBM

   To assess the role of PTEN in modulating macrophage chemoattraction and
   polarization, we conducted a series of functional studies in
   PTEN–wild-type (PTEN-WT) and PTEN-deleted/mutated (PTEN-Del/Mut) glioma
   models. First, we evaluated PTEN expression in 23 GBM cell lines by
   immunoblot analysis and found that PTEN expression was undetectable in
   U118, U87, U138, U251, SYU354, SF295, U373, and MGR3 cell lines
   ([68]Fig. 1A). By using Sanger sequencing technology, we confirmed the
   mutation status of PTEN gene in all glioma cell lines used in our study
   (table S1). We cultured the 23 GBM cell lines in fetal bovine serum
   (FBS)–free medium for 24 hours and collected the supernatants for use
   as glioma cell–derived conditioned medium (GCM) in subsequent
   experiments.

Fig. 1. PTEN deficiency facilitates macrophage infiltration and M2
polarization.

   [69]Fig. 1.
   [70]Open in a new tab

   (A) Immunoblots for PTEN in 23 glioma cell lines. (B) Quantification of
   relative migration of THP-1 macrophages following stimulation with CM
   from eight PTEN-null cell lines (U118, U87, U138, U251, SYU354, SF295,
   U373, and MGR3) and eight PTEN–wild-type (WT) GBM cell lines (SKMG-1,
   LN229, SYU687, MGR2, LNZ308, GL261, SHG44, and T98). n = 3 biological
   replicates. (C) Immunoblots showing PTEN expression status in
   PTEN-deleted (Del) U87/U118 cells after PTEN overexpression (OE) and
   PTEN-WT SKMG-1/LN229/GL261 cells after PTEN knockout (KO). (D and E)
   Flow cytometry analysis of CD163 expression in BMDMs treated with CM
   from PTEN-Del/OE U87 cells and PTEN-WT/KO GL261 cells. (F and G)
   Immunofluorescence staining of CD86 and CD163 was determined and
   quantified in intracranial PTEN-WT/KO GL261 and PTEN-Del/OE U87
   xenograft tumors. PTEN-WT/KO GL261 xenografts were established in
   C57BL/6J mice, and PTEN-Del/OE U87 xenografts were established in
   BALB/c-nu/nu mice. Scale bar, 200 μm. **P < 0.01 and ***P < 0.001.

   Transwell migration assays showed that more macrophages migrated in GCM
   from eight PTEN-Del GBM cell lines, compared with GCM from eight
   PTEN-WT GBM cell lines ([71]Fig. 1B and fig. S1A). We then knocked out
   the PTEN gene (PTEN-KO) in the PTEN-WT cell lines SKMG-1, LN229, and
   GL261 using CRISPR technology and overexpressed WT PTEN (PTEN-OE) in
   the PTEN-Del cell lines U87 and U118 ([72]Fig. 1C). Transwell migration
   assays using these cell lines confirmed that GCM from PTEN-KO SKMG-1,
   LN229, and GL261 cells greatly enhanced macrophage migration compared
   with GCM of the parental PTEN-WT cells, while GCM from U87 and U118
   (PTEN-Del) cells reduced macrophage migration after WT PTEN restoration
   (fig. S1, B and C).

   To figure out how PTEN affected macrophage polarization, flow cytometry
   and quantitative real-time polymerase chain reaction (qRT-PCR) were
   used to evaluate M1 and M2 markers in macrophages treated with GCM. As
   compared with GCM from PTEN-normal cells (GL261-WT and U87-OE), the
   expression level of M2 marker CD163 on bone marrow–derived macrophages
   (BMDMs) was enhanced by GCM from PTEN-deficient cells (GL261-KO and
   U87-Del) ([73]Fig. 1, D and E, and fig. S1D). In addition, the
   expression level of the M1 marker CD86 on BMDM and THP-1 macrophages
   was down-regulated by GCM derived from PTEN-deficient cells (fig. S1E).
   The qRT-PCR results indicated that GCM obtained from PTEN-normal glioma
   cells seemed to induce higher expression of M1 markers (IL1, IL12,
   IL23, and TNF-α) in macrophages, compared with GCM obtained from
   PTEN-deficient cells (fig. S1F). Consistently, compared to PTEN-normal
   xenografts (GL261-WT and U87-OE) grown in vivo, PTEN-deficient
   xenografts (GL261-KO and U87-Del) showed faster growth in subcutaneous
   mouse models, shortened the survival time of orthotopic mouse models
   (fig. S1, G and H), and enhanced macrophage M2 polarization with higher
   CD163 and lower CD86 expression scores ([74]Fig. 1, F and G). Together,
   these data demonstrate that PTEN-deficient GBM cells facilitate
   macrophage infiltration and promote macrophage M2 polarization more
   efficiently than PTEN-WT GBM cells.

Gal-9 is preferentially secreted by PTEN-deficient glioma cells and promotes
macrophage M2 polarization

   To identify the secreted proteins that modulated macrophage M2
   polarization in PTEN-deficient GBM, we examined putative
   macrophage-recruiting factors and macrophage polarization–related
   factors exhibiting a >1-fold change in PTEN-KO LN229 cells according to
   a secreted protein database ([75]18) (table S2). Transcriptome and
   qRT-PCR analyses of LN229-WT and LN229-KO cells showed increased
   expression in PTEN-KO cells for LGALS9, IL6, VEGFC, SPP1, TGFB1, and
   TGFB2, with LGALS9 [which encodes the galectin-9 (Gal-9) protein]
   showing a marked increase ([76]Fig. 2, A and B). Similarly, upon PTEN
   deletion, LGALS9 mRNA levels were increased to the greatest extent
   among these six genes in other PTEN-WT GL261 and SKMG-1 cells (fig.
   S2A). We further observed that genes encoding macrophage
   polarization–related factors (LGALS9, IL6, TGFB1, and TGFB2) were
   down-regulated upon PTEN rescue in U87 cells, with the LGALS9 gene
   showing a marked decrease (fig. S2B). Moreover, we used Quantibody
   Human Cytokine Antibody Array 440 to perform a more comprehensive
   proteomic analysis of the CM from LN229-KO and LN229-WT cells. We
   screened out the top 22 factors among the 440 secreted proteins
   evaluated in the array (LN229-KO CM/LN229-WT CM > 3) (fig. S2C). Next,
   among the 22 factors, 5 macrophage M2 polarization–related factors
   {Gal-9, macrophage migration inhibitory factor (MIF),
   latency-associated peptide (LAP) [transforming growth factor–β1
   (TGF-β1)], brain-derived neurotrophic factor (BDNF), and basic
   fibroblast growth factor (bFGF)} were chosen for further validation. We
   found that human recombinant Gal-9 protein was much easier than other
   four cytokines to induce macrophage M2 polarization, within higher
   production of interleukin-10 (IL-10) and TGF-β1 by THP-1 macrophages
   (fig. S2D). To further confirm how PTEN deficiency affects the cytokine
   profiles of GBM cells, we compared the cytokines secreted by
   PTEN-deficient and PTEN-normal GBM cells (PTEN-WT versus KO SKMG-1
   cells; PTEN-Del versus OE U87 cells) using antibody microarrays.
   Compared with PTEN-normal cells, we noticed that PTEN deficiency did
   not cause notable changes in the expression of some macrophage
   polarization–related factors [interferon-γ (IFN-γ), tumor necrosis
   factor–α (TNF-α), IL-4, IL-10, TGF-β, etc.] by GBM cells (U87 and
   SKMG-1 KO cells) ([77]Fig. 2C and fig. S2E). Unexpectedly,
   enzyme-linked immunosorbent assay (ELISA) results confirmed that the
   GCM of PTEN-KO or PTEN-Del cells showed markedly high Gal-9 protein
   levels, while the GCM of PTEN-WT or PTEN-OE cells exhibited low levels
   ([78]Fig. 2D). Consistently, LGALS9 mRNA levels were up-regulated upon
   CRISPR-mediated deletion of PTEN in PTEN-WT GL261, SKMG-1, and LN229
   cells, and it was down-regulated by reexpression of PTEN in PTEN-Del
   U87 and U118 cells (fig. S2F). Moreover, by analyzing with Cancer Cell
   Line Encyclopedia datasets, we noticed that the LGALS9 mRNA levels of
   25 PTEN-Del/Mut GBM cell lines were significantly higher than those of
   26 PTEN-WT GBM cell lines (fig. S2G), which was similar to the Gal-9
   secretion pattern in the GBM cell lines cultured in our laboratory
   (fig. S2H). The relationship between Gal-9 and PTEN was ultimately
   confirmed by immunoblotting ([79]Fig. 2E) and further reinforced by the
   positive correlation between phospho-AKT and Gal-9 in the indicated
   glioma cell lines ([80]Fig. 2F). In addition, eight patient-derived
   short-term glioma stem cell lines (GSCs) were used to verify the
   up-regulation of Gal-9 in PTEN-negative GSCs ([81]Fig. 2G).

Fig. 2. Gal-9, a modulator of macrophage M2 polarization, is abundantly
secreted by PTEN-deficient glioma cells.

   [82]Fig. 2.
   [83]Open in a new tab

   (A) Transcriptome analyses of up-regulated genes encoding secreted
   proteins in PTEN-KO LN229 cells. Values are expressed as the fold
   change between PTEN-KO and PTEN-WT LN229 cells. The top 11 genes (with
   fold changes of >2.4) were chosen for further validation. (B) qRT-PCR
   validation of the 11 genes and the TNF gene in PTEN-KO/WT LN229 cells.
   Values are expressed as the fold change between PTEN-KO and PTEN-WT
   LN229 cells after normalization to the housekeeping gene GAPDH. (C)
   Macrophage-inducible factors in PTEN-KO/WT SKMG-1 cells and PTEN-Del/OE
   U87 cells were detected using antibody microarrays. (D) ELISA analysis
   of GCM showed Gal-9 secretion up-regulated by PTEN-KO in LN229/SKMG-1
   cells and down-regulated by PTEN-OE in U87/U118 cells. **P < 0.01 and
   ***P < 0.001, Student’s t test. (E and F) Immunoblots of Gal-9 and PTEN
   or p-AKT expression in the indicated glioma cell lines. (G) Immunoblots
   for PTEN and Gal-9 in cell lysates of eight short-term patient-derived
   glioma stem cell lines (GSCs): GSC-1, GSC-#363, GSC-#624, GSC-11,
   GSC-#354, GSC-#687, GSC-#481, and GSC-#530. (H) Flow cytometry analysis
   of CD163 and CD206 on THP-1 macrophages treated with U87 CM or
   recombinant Gal-9 protein (50 ng/ml). (I) Immunoblots showing
   successful knockdown of Gal-9 in U87 cells by shGal-9. (J) Flow
   cytometry analysis of CD163 and CD206 on THP-1 macrophages treated with
   U87 CM or U87 shGal-9#1 CM. (K) Flow cytometry analysis of CD163 and
   CD206 on THP-1 macrophages treated with U87 shGal-9#1 CM or U87
   shGal-9#1 CM supplemented with α-lactose (40 μM).

   To validate the capacity of the Gal-9 protein to function as a
   modulator of macrophage M2 polarization in vitro, we performed flow
   cytometry and short hairpin RNA (shRNA)–mediated Gal-9 knockdown
   experiments and observed that recombinant human Gal-9
   protein–supplemented medium induced M2 marker CD163 and CD206
   expression in macrophages, demonstrating the same effect as U87
   cell-derived CM ([84]Fig. 2H and fig. S3A). Together, CM obtained from
   PTEN-Del U87 cells expressing Gal-9 shRNA ([85]Fig. 2I) significantly
   inhibited M2 markers CD163 and CD206 expression levels ([86]Fig. 2J and
   fig. S3B), and also inhibited macrophage migration rate (fig. S3C),
   compared with the control GCM. In addition, Gal-9 concentrations in the
   CM from 16 GBM cell lines exhibited a strong positive correlation with
   the number of migrating macrophages (fig. S3D), and elevated
   concentrations of human recombinant Gal-9 protein obviously increased
   the migration rate of macrophages (fig. S3E). α-Lactose (40 μM), a
   Gal-9 inactivator, negligibly inhibited CD163 and CD206 expression
   further in CM derived from U87 cells expressing Gal-9 shRNA ([87]Fig.
   2K and fig. S3F). In general, these results suggest that up-regulated
   Gal-9 expression in PTEN-deficient GBM cells is strongly associated
   with macrophage M2 polarization and macrophage migration.

Gal-9 is up-regulated by PTEN deficiency via the AKT-GSK3β-IRF1 pathway in
glioma cells

   To explore PTEN regulation of Gal-9 protein expression, a Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis
   was performed to catalog signaling pathways modulated by PTEN deletion.
   We found that the phosphatidylinositol 3-kinase (PI3K)–AKT signaling
   pathway was the top pathway activated in PTEN-KO LN229 cells (fig.
   S4A), which was consistent with other studies ([88]6), reporting that
   AKT, SRC, NOTCH1, and YAP1 pathways were enriched in PTEN-null glioma
   cells compared with PTEN-WT controls. To investigate the relevance of
   these pathways in regulating Gal-9, PTEN-KO LN229 cells were treated
   with various pathway inhibitors, including the AKT inhibitor MK2206,
   SRC inhibitor KX2-391, NOTCH1 inhibitor LY3039478, and YAP1 inhibitor
   verteporfin. In PTEN-KO LN229 cells, the AKT inhibitor, rather than
   other inhibitors, significantly inhibited Gal-9 protein expression and
   mRNA expression (fig. S4, B and C). On the other hand, the AKT
   activator SC 79 up-regulated AKT phosphorylation, and then Gal-9
   expression increased in PTEN-WT cells (fig. S4D). Moreover, the AKT
   inhibitor suppressed Gal-9 expression in dose- and time-dependent
   manners in PTEN-KO cells ([89]Fig. 3, A and B). Glycogen synthase
   kinase 3β (GSK3β) and mammalian target of rapamycin (mTOR) are two
   downstream molecules of the AKT pathway, and we noticed that the GSK3β
   inhibitor tideglusib up-regulated Gal-9 levels in PTEN-KO SKMG-1 cells
   in a dose-dependent manner, while the mTOR inhibitor rapamycin had no
   regulatory effect on Gal-9 expression (fig. S4E). AKT activator SC 79
   could inhibit GSK3β activity by up-regulating p-GSK3β^Ser9 levels in
   PTEN-WT cells (fig. S4F). In contrast, the AKT inhibitor MK2206 reduced
   p-GSK3β^Ser9 levels ([90]Fig. 3, A and B) and up-regulated
   p-GSK3β^Try216 levels (fig. S4G) in PTEN-KO glioma cells, resulting in
   the switch of more GSK3β to its active state. We also observed that
   PTEN-KO cells exhibited higher levels of p-GSK-3β^Ser9 and thus
   repression of GSK3β activity compared to PTEN-WT cells (fig. S4H).
   Therefore, the inactivation of GSK3β mediated by the AKT pathway may
   contribute to Gal-9 up-regulation.

Fig. 3. PTEN deficiency–induced Gal-9 production is regulated via the
AKT-GSK3β-IRF1 pathway.

   [91]Fig. 3.
   [92]Open in a new tab

   (A and B) Immunoblots of Gal-9, AKT, p-AKT^Ser473, and p-GSK3β^Ser9 in
   PTEN-KO LN229/SKMG-1 cells treated with the AKTi MK2206 at different
   concentrations and time points. (C and D) Immunoblots of p-GSK3β^Ser9,
   GSK3β, p-GSK3α^Tyr279, p-GSK3β^Tyr216, IRF1, and Gal-9 in PTEN-KO LN229
   cells treated with tideglusib at different concentrations and for
   different time lengths. (E) Immunoblots of IRF1 and lamin B1 in the
   cytoplasmic and nuclear components of PTEN-KO LN229 cells treated with
   different concentrations of tideglusib. (F) Immunoblots of Gal-9 and
   IRF1 in PTEN-KO LN229/SKMG-1 cells transfected with siIRF1 or control
   siRNA (siNC). (G) A dual-luciferase reporter system containing the
   2.0/1.5/0.5-kb human LGALS-9 promoter DNA fragments was used to predict
   their transcription activity in indicated glioma cells. The values are
   expressed as the ratio of Firefly and Renilla luciferase signals. ***P
   < 0.001, one-way ANOVA. (H) The transcriptional activity of the 0.5-kb
   human LGALS-9 promoter DNA fragment was detected in PTEN-KO
   LN229/SKMG-1 cells transfected with siIRF1 or siNC. ***P < 0.001,
   one-way ANOVA. Band density of immunoblots was determined by ImageJ
   software, and values represent the expression levels after
   normalization to the density of GADPH/β-tubulin.

   To determine how the AKT pathway transcriptionally regulates Gal-9
   expression, we used the USCS genome browser to predict the potential
   transcription factors (TFs) that bind to the promoter region of the
   LGALS-9 gene (fig. S5A). The top five TFs with the greatest
   promoter-binding potential—EHF, ZNF460, Sox11, IRF1, and ZNF684—were
   chosen to determine their putative binding sites in the promoter region
   of LGALS-9 using an in silico promoter region screening approach
   (JASPAR) (fig. S5B). Among the five TFs, only IRF1 was specifically
   highly expressed in mesenchymal subtype GBM in the Chinese Glioma
   Genome Atlas (CGGA) database (fig. S5C). qRT-PCR results indicated that
   only IRF1 mRNA levels were significantly up-regulated by PTEN-KO in
   LN229 cells, and EHF, ZNF460, IRF1, and ZNF684 were all reduced by
   PTEN-OE in U87 cells (fig. S5D). LGALS-9 expression positively
   correlated with IRF1 expression in the CGGA (fig. S5E) and Gene
   Expression Profiling Interactive Analysis (GEPIA) glioma samples (fig.
   S5F).

   Then, we focused on IRF1 as the most possible TF regulating LGALS-9
   gene expression and conducted immunoblot analysis of Gal-9 and IRF1 in
   PTEN-WT or KO LN229 and SKMG-1 cells treated with GSK3β inhibitors.
   Previous studies have validated that active GSK3β promotes IRF1
   degradation, which led to less IRF1 in cytoplasm entering the nucleus
   ([93]19). In our study, it showed that the expression levels of IRF1
   and Gal-9 were increased by tideglusib in both LN229 KO cells ([94]Fig.
   3, C and D) and SKMG-1 WT cells (fig. S6A) in a dose- or time-dependent
   manner upon up-regulation of p-GSK3β^Ser9. We also used another GSK3β
   inhibitor, LiCL, to examine whether it could phenocopy the effect of
   AKT activator SC 79 on Gal-9 and IRF1 expression. The results showed
   that LiCL could up-regulate the levels of p-GSK3β^Ser9, IRF1, and Gal-9
   in both PTEN-KO and PTEN-WT cells (fig. S6B). In addition, the AKT
   activator SC 79 could also up-regulate IRF1 levels, similar with
   PTEN-KO effect in PTEN-WT cells (fig. S6C). Furthermore, IRF1 protein
   expression was up-regulated by tideglusib in the nuclei of PTEN-KO
   LN229 cells in a dose-dependent manner ([95]Fig. 3E). Therefore, high
   levels of the inactive GSK3β (p-GSK3β^Ser9) contributed to Gal-9 and
   IRF1 up-regulation.

   To assess the role of IRF1 in regulating Gal-9 expression directly, we
   silenced IRF1 using small interfering RNA (siRNA) in PTEN-deficient GBM
   cells, which showed marked suppression of Gal-9 expression, compared
   with the control group ([96]Fig. 3F). To identify the promoter region
   involved in IRF1-mediated regulation of LGALS-9 expression, we used a
   series of dual-luciferase reporter vectors containing various fragments
   of the human LGALS-9 promoter (2.0/1.5/0.5 kb). In PTEN-KO LN229/SKMG-1
   cells, the Firefly luciferase activity driven by the 0.5-kb fragment
   was higher than that of the 1.5-kb fragment; the full-length 2.0-kb
   fragment showed a lower ratio of Firefly/Renilla luciferase signals,
   perhaps because of low transfection efficiency ([97]Fig. 3G). Moreover,
   the Firefly/Renilla values of the 0.5-kb fragment were significantly
   down-regulated by PTEN expression rescue in U87 cells ([98]Fig. 3G).
   Further, si-IRF1 obviously inhibited the expression of the Firefly
   luciferase reporter (0.5 kb), suggesting that IRF1 truly enhanced
   LGALS-9 transcription via binding to its promoter at 0.5-kb promoter
   region ([99]Fig. 3H), consistent with the predicted binding sites by
   JASPAR. Together, the above results indicate that Gal-9 production in
   PTEN-null glioma cells is modulated through enhanced AKT activation
   that suppresses IRF1 degradation by phosphorylating GSK3β into inactive
   form, leading to maintained IRF1 activation and thereby Gal-9
   transcription.

Multilayer signaling network reveals Gal-9/Tim-3–mediated interactions
between glioma cells and macrophages

   To further figure out how glioma-secreted Gal-9 promoted macrophage
   polarization, we leveraged the single-cell RNA sequencing (scRNA-seq)
   data of human GBM tissues from three patients (PTEN-null: M47; PTEN-WT:
   M48 and M58) (fig. S7, A and B) to infer cell-cell interactions. The
   cells were clustered and annotated into seven types ([100]Fig.
   4A)—glioma cells (malignant), macrophages, oligodendrocytes, T cells,
   fibroblasts, neutrophils, and endothelial cells—according to marker
   gene expression (fig. S8) and copy number variation scores (fig. S9).
   The fraction of macrophages in the PTEN-null sample was larger than
   that in the PTEN-WT samples (fig. S10), consistent with the
   aforementioned results ([101]Fig. 1B). On the basis of the scRNA-seq
   data, we developed a multilayer network approach to infer inter- and
   intracellular signaling networks (see Materials and Methods). Cell-cell
   communication networks ([102]Fig. 4B) globally depicted the
   intercellular ligand-receptor (LR) interactions between the above cell
   types. Abundant LR interactions between malignant glioma cells and
   macrophages were observed in both PTEN-null and PTEN-WT samples.
   However, compared to the PTEN-WT samples, Gal-9 was found to interact
   with T cell immunoglobulin and mucin domain 3 (Tim-3), a receptor of
   macrophages, specifically in the PTEN-null sample that expressed high
   levels of Gal-9 and Tim-3 ([103]Fig. 4C and fig. S7B). Moreover, the
   multilayer network analysis ([104]Fig. 4D) revealed that Gal-9/Tim-3
   signaling activated several TFs (e.g., STAT1, ESR1, and ELK1) in
   macrophages, and the downstream target genes involved many marker genes
   related to macrophage polarization [e.g., HMOX1, PPARD, CEBPA, and
   ADCY7 genes ([105]16)].

Fig. 4. scRNA-seq data analysis reveals Gal-9/Tim-3–mediated interactions
between glioma cells and macrophages.

   [106]Fig. 4.
   [107]Open in a new tab

   (A) Distribution of single cells’ gene expression colored by cell types
   or samples. On the basis of the expression of marker genes (fig. S8),
   seven cell types were identified. (B) Cell-cell communication networks
   for PTEN-null and PTEN-WT samples. The color of the dot represents the
   cell type, the edge represents the communication between the two cell
   types whose thickness represents the number of downstream target genes
   of upstream activated ligand-receptor (LR) pairs, and color represents
   the direction of cell communication (e.g., the purple edge represents
   the signals delivered by malignant glioma cells). (C) Stacked violin
   plots of gene expressions of Gal-9 and Tim-3 in two groups of samples
   (PTEN-null and PTEN-WT). (D) Gal-9/Tim-3 interaction between glioma
   cells and macrophages inferred by the multilayer network revealed that
   macrophage M2 marker genes were regulated by Gal-9/Tim-3 signaling
   through the downstream pathways and transcription factors (TFs) (e.g.,
   ESR1, ELK1, and STAT1). (E) GO enrichment analysis for the Gal-9/Tim-3
   downstream target genes. (F) GSEA enrichment analysis for the ranked
   gene lists in macrophages according to fold changes of gene expression
   in the PTEN-null sample compared with PTEN-WT samples.

   Furthermore, we performed functional enrichment analysis to investigate
   the functions of the downstream target genes of Gal-9/Tim-3 in the
   multilayer network. The Gal-9/Tim-3–regulated target genes were
   significantly enriched in Gene Ontology (GO) biological processes
   including macrophage activation (polarization) or vascular growth and
   development ([108]Fig. 4E). Consistently, gene set enrichment analysis
   (GSEA), a functional class scoring analysis, showed that highly
   expressed genes in macrophages in the PTEN-null sample relative to the
   PTEN-WT samples were positively correlated with the gene sets related
   to macrophage activation, as well as vascular growth and development
   ([109]Fig. 4F). These results imply that Gal-9/Tim-3–mediated
   glioma-macrophage interactions play an important role in macrophage
   polarization and angiogenesis.

Enhanced Gal-9/Tim-3 signaling predicts poor prognosis in glioma patients

   To ascertain the clinical significance of Gal-9 and its receptor Tim-3
   in glioma, we evaluated Gal-9 and Tim-3 expression in human glioma
   tissues. In the CGGA database, Gal-9 and Tim-3 expressions were
   analyzed in four GBM subtypes, and we found that Gal-9 and Tim-3
   expression in the mesenchymal GBM subtype was the highest among the
   four subtypes (fig. S11A). Patients with mesenchymal subtype GBM
   exhibited significantly poorer survival than those with nonmesenchymal
   subtype GBM (fig. S11B). We also observed positive correlations of
   Tim-3, Gal-9, and CD68 expression, suggesting the importance of
   Gal-9/Tim-3 signaling up-regulation in glioma macrophages (fig. S11C).
   The GEPIA database demonstrated that Gal-9 expression in GBM and
   low-grade glioma (LGG) was significantly higher than in normal tissue
   (fig. S11D). Notably, glioma patients with high Gal-9 or (and) Tim-3
   expression showed shorter survival in the CGGA dataset ([110]Fig. 5A),
   TCGA dataset ([111]Fig. 5B), and [112]GSE16011 dataset ([113]Fig. 5C),
   respectively. Moreover, tyramide signal amplification
   immunohistochemistry (TSA-IHC) staining verified that Gal-9 and CD163
   were expressed at higher levels in human PTEN-Del GBM tissues than in
   PTEN-WT GBM tissues ([114]Fig. 5D). We also examined 240 human glioma
   samples stained for Gal-9 and Tim-3 using TSA-IHC ([115]Fig. 5E), and
   staining intensity was quantified using the inForm System. Kaplan-Meier
   analysis revealed that high expression scores for Gal-9/Tim-3,
   individually or together, predicted shorter survival of GBM patients
   than those with low Gal-9/Tim-3 features (P < 0.0001; [116]Fig. 5F).
   Taken together, these results indicate that Gal-9/Tim-3 signaling
   strongly correlates with the mesenchymal GBM subtype and predicts a
   poor prognosis of glioma patients.

Fig. 5. High activation of Gal-9/Tim-3 signaling predicts poor survival of
human glioma patients.

   [117]Fig. 5.
   [118]Open in a new tab

   Kaplan-Meier survival curves of glioma patients stratified by Gal-9
   expression, Tim-3 expression, or the product of Gal-9 and Tim-3
   expression (Gal-9 × Tim-3) in the CGGA dataset (A), TCGA dataset (B),
   and [119]GSE16011 dataset (C). (D) Immunofluorescence staining and
   quantification of Gal-9 and CD163 expression in human GBM tissues with
   WT (n = 9) or deleted (n = 9) PTEN gene. Scale bar, 100 μm. **P < 0.01
   and ***P < 0.001, Student’s t test. (E) Representative
   immunofluorescence staining for Gal-9 and Tim-3 in human glioma tissue
   microarrays. Scale bar, 200 μm. (F) Kaplan-Meier survival curves
   indicate that high Gal-9 and/or Tim-3 expression correlates with a poor
   prognosis for glioma patients. Differences in Kaplan-Meier curves were
   assessed using the log-rank test.

Blockade of Gal-9/Tim-3 signaling inhibits macrophage M2 polarization,
suppresses tumor growth, and prolongs survival in GBM models

   To validate the role of Gal-9/Tim-3 signaling in modulating macrophage
   M2 polarization, we used both pharmacologic (α-lactose/anti–Tim-3
   antibody) and genetic (shRNAs) interference methods in PTEN-deficient
   glioma cells. Immunofluorescence showed that GCM supplemented with
   α-lactose markedly reduced Gal-9 protein binding to macrophage surface
   in vitro (fig. S12A). PTEN-deficient GCM supplemented with different
   concentrations of α-lactose was used to treat THP-1 macrophages for 72
   hours. Further, flow cytometry results showed that α-lactose had a
   dose-dependent effect on reducing M2 polarization, as indicated by the
   reduced expression of CD163 and CD206 markers in THP-1 macrophages
   ([120]Fig. 6A and fig. S12B). The qRT-PCR results showed that GCM
   supplemented with α-lactose induced significant up-regulation of the M1
   markers (TNF-α, IL-12, and IL-23) and down-regulation of M2 markers
   (IL-10 and TGF-β1) to some extent in THP-1 macrophages and BMDMs (fig.
   S12C). In contrast to GCM-treated macrophages, GCM mixed with
   anti–Tim-3 antibody or α-lactose increased the expression level of
   macrophage M1 marker CD86, as detected by flow cytometry ([121]Fig. 6B
   and fig. S12D), and the expression level of IL-1β, as detected by
   cytokine microarray analysis (fig. S12E). Notably, neither Gal-9
   knockdown (fig. S12F) nor α-lactose/anti–Tim-3 antibody (fig. S12G)
   affected the proliferation of PTEN-deficient glioma cells in vitro.

Fig. 6. Blockade of Gal-9/Tim-3 signaling inhibits macrophage M2 polarization
and GBM growth in vivo.

   [122]Fig. 6.
   [123]Open in a new tab

   (A) Flow cytometry analysis of CD163 and CD206 on THP-1 macrophages
   treated with different doses of α-lactose–supplemented GCM. (B) Flow
   cytometry analysis of CD86 on BMDMs or THP-1 macrophages treated with
   GCM, GCM + anti–Tim-3 antibody (1 μg/ml), or GCM + α-lactose (40 μM).
   (C to E) Survival curves of BALB/c-nu/nu mice bearing orthotopic U87
   xenografts (C), C57BL/6J mice bearing orthotopic GL261 KO xenografts
   (D), and Wistar rats bearing orthotopic C6 xenografts (E) treated with
   α-lactose or anti–Tim-3 antibody. (F) Gal-9 knockdown prolonged the
   survival of BALB/c-nu/nu mice bearing orthotopic U87 xenografts. (G)
   Representative immunofluorescence staining and quantification of
   Gal-9/CD163 in orthotopic PTEN-Del U87 and PTEN-KO GL261 tumors treated
   with α-lactose or anti–Tim-3 antibody. Scale bar, 200 μm. ns, no
   significance. ***P < 0.001, one-way ANOVA. (H) Representative
   immunofluorescence staining and quantification of Gal-9/CD163 in
   orthotopic and subcutaneous U87-shC and U87-shGal-9 tumors. Scale bar,
   200 μm. Tumors were harvested on day 22 after implantation. n = 3
   biological replicates. ***P < 0.001, Student’s t test. In (C) to (F),
   *P < 0.05, **P < 0.01, and ***P < 0.001, log-rank test.

   We next assessed the effect of Gal-9/Tim-3 blockade on tumor growth and
   macrophage polarization in vivo. We found that α-lactose or anti–Tim-3
   antibody significantly delayed tumor growth (fig. S13, A and B) and
   extended the survival of both orthotopic PTEN-null U87 ([124]Fig. 6C)
   and PTEN-KO GL261 ([125]Fig. 6D) mouse glioma models, and an orthotopic
   PTEN-null C6 rat glioma model ([126]Fig. 6E). Unexpectedly,
   shRNA-mediated depletion of Gal-9 expression markedly slowed tumor
   growth in subcutaneous and orthotopic PTEN-null U87 GBM models (fig.
   S13, C to E) and extended the survival of mice bearing orthotopic U87
   xenografts ([127]Fig. 6F). Accordingly, CD163^+ macrophage infiltration
   was markedly reduced in GBM xenografts by α-lactose or anti–Tim-3
   antibody treatments ([128]Fig. 6G), or by shRNA-mediated Gal-9
   knockdown ([129]Fig. 6H) in GBM mouse models, in accordance with the
   down-regulated Gal-9 expression induced by α-lactose or shGal-9. In
   orthotopic C6 xenografts, the expression of the M1 macrophage marker
   CD86 was accordingly up-regulated in the α-lactose and anti–Tim-3
   antibody groups (fig. S13F).

α-Lactose inhibits TAM-derived VEGFA expression levels and attenuates
angiogenesis in glioma

   We noticed that inhibition of Gal-9 expression by α-lactose or shRNA
   effectively impeded glioma growth in vivo but not in vitro, prompting
   speculation that TAMs may secrete factors supporting glioma growth and
   angiogenesis. Our computational analysis ([130]Fig. 4, E and F) implied
   that the downstream target genes of Gal-9/Tim-3 signaling were
   significantly enriched in VEGF production and vasculature development.
   Therefore, we examined tumor angiogenesis to understand how
   α-lactose–regulated TAMs inhibit GBM progression. We found that blood
   vessels were reduced in response to α-lactose in subcutaneous
   ([131]Fig. 7A) and orthotopic ([132]Fig. 7B) PTEN-KO GL261 xenografts.
   In subcutaneous PTEN-KO GL261 tumors treated with or without α-lactose
   (fig. S14A), IHC analysis (fig. S14B) indicated that CD31 expression
   (vessel density) positively correlated with tumor size (fig. S14C),
   Gal-9 expression (fig. S14D), and CD163 expression (fig. S14E). GSEA
   and RNA-seq analysis revealed that VEGFA signaling was highly
   activated, and 50 angiogenesis-related genes were up-regulated in
   GCM-treated monocytes, compared with those in untreated monocytes
   ([133]Fig. 7, C and D). GSEA for two VEGFA signatures in TCGA samples
   suggested a highly activated VEGFA pathway in the GBM samples with
   macrophage M2–positive correlation ([134]Fig. 7E), and VEGFA was
   positively correlated with the M2 macrophage marker CD163 in the CGGA
   database ([135]Fig. 7F). Furthermore, qRT-PCR and immunoblots
   demonstrated that GCM supplemented with α-lactose significantly
   inhibited VEGFA expression in GCM-treated THP-1 macrophages and BMDMs
   ([136]Fig. 7, G and H). Thus, these data suggest that α-lactose can
   decrease VEGFA expression that was initially up-regulated by TAMs,
   thereby inhibiting glioma angiogenesis. A schematic model that
   summarizes results of our study is shown in [137]Fig. 7I: (i) PTEN
   deficiency in GBM cells promotes macrophage M2 polarization via the
   intercellular Gal-9/Tim-3 interaction, with Gal-9 secretion from glioma
   cells directly modulated by the AKT-GSK3β-IRF1 pathway; (ii)
   Gal-9/Tim-3 signaling can be blocked by α-lactose or anti–Tim-3
   antibody; and (iii) Gal-9/Tim-3 inhibition impairs PTEN-deficient GBM
   progression by decreasing TAM-derived VEGFA, thus attenuating
   angiogenesis.

Fig. 7. α-Lactose inhibits glioma angiogenesis by impairing TAM-derived VEGFA
expression.

   [138]Fig. 7.
   [139]Open in a new tab

   (A) Representative images of subcutaneous PTEN-KO GL261 tumors treated
   with or without α-lactose. (B) IHC showing CD31^+ vessel density in
   intracranial PTEN-KO GL261 tumors treated with or without α-lactose.
   Scale bar, 200 μm. n = 3 biological replicates. (C and D) GSEA for
   VEGFA signatures and heatmap representation of 50 up-regulated VEGFA
   pathway–associated genes in healthy monocytes treated with or without
   GCM. (E) GSEA for two VEGFA signatures in M2 macrophage–positive versus
   M2 macrophage–negative correlated TCGA GBM samples. NES, normalized
   enrichment score. (F) VEGFA positively correlates with CD163 in the
   CGGA database. (G) qRT-PCR and (H) immunoblot validation of VEGFA
   expression in THP-1 macrophages and BMDMs treated with GCM or GCM +
   α-lactose. ***P < 0.001, Student’s t test. (I) Schematic model of the
   mechanism underlying how glioma cells with PTEN deficiency regulate
   macrophage polarization through Gal-9/Tim-3 signaling, which can be
   blocked by α-lactose or anti–Tim-3 antibody.

DISCUSSION

   Surgical resection followed by adjuvant radiochemotherapy is the
   standard treatment for GBM and produces only modest benefits for
   survival ([140]20). Immunotherapy with checkpoint inhibitors has
   revolutionized the treatment of a variety of tumors, but only a small
   subset of GBM patients (<10%) shows objective responses to
   immunotherapy ([141]21). In this study, we explored the mechanisms
   underlying the polarization of TAMs, the most abundant nontumor cells
   in GBM tumors ([142]4), and assessed their roles in angiogenesis in
   PTEN-null GBM. Previous studies have shown that high stromal and immune
   signatures are enriched in mesenchymal GBM and are specifically
   correlated with genetic alterations of the PTEN-PI3K pathway ([143]6).
   These findings suggested that genetic driver mutations such as PTEN
   alteration can create unique TMEs with abundant macrophage infiltration
   ([144]2, [145]6). Here, we confirmed that PTEN-null glioma cells
   recruited more macrophages than PTEN-WT glioma cells, consistent with
   other studies ([146]6). First, we found that these macrophages tended
   to be polarized into M2 macrophages when cocultured with GCM, and
   PTEN-deficient glioma cells were more effective in promoting macrophage
   M2 polarization than PTEN-WT glioma cells. Subtype-specific alterations
   like PTEN deficiency can provide promising therapeutic targets in GBM.
   Our results indicate that macrophage-centered approaches may be
   effective for treating PTEN-null glioma but may not be effective for
   glioma with the WT PTEN gene.

   To mimic the effect of PTEN gene alterations on the TME in glioma, we
   knocked out PTEN in LN229, SKMG-1, and GL261 cells, and overexpressed
   WT PTEN in U87 and U118 cells. Erasing PTEN protein in PTEN-WT cells
   using CRISPR technology could simulate one kind of PTEN gene deficiency
   causing PTEN protein loss, similar to the natural loss of PTEN protein
   in PTEN-Del cells. We examined the functions of both PTEN-OE and
   PTEN-WT cell lines in modulating macrophage polarization. In response
   to PTEN overexpression, the mRNA levels of LGALS9 in LN229-KO cells
   were restored to levels observed in LN229-WT cells (fig. S15A). There
   was no significant difference in the modulation of macrophage M2
   polarization among LN229-WT cells, PTEN-overexpressing LN229-KO cells
   (LN229 KO + OE), and PTEN-overexpressing LN229-WT cells (LN229 WT + OE)
   (fig. S15B). These results suggested that our cell models were suitable
   to explore how PTEN loss in glioma affected macrophage polarization.

   We then found that PTEN deficiency–mediated high expression of Gal-9
   was involved in macrophage M2 polarization. Furthermore, combined
   profiling and functional experiments revealed that PTEN deficiency in
   glioma cells promoted Gal-9 expression and secretion. Gal-9, a
   β-galactoside–binding protein implicated in modulating cell-cell and
   cell-matrix interactions, is inhibited by α-lactose ([147]22). Another
   report showed that Gal-9 promoted M2 polarization by activating Tim-3
   receptors on BMDMs and that α-lactose blocked Gal-9/Tim-3 signaling
   ([148]22). Therefore, in this study, α-lactose and anti–Tim-3 antibody
   were used to block Gal-9/Tim-3 signaling in vitro and in vivo, and both
   treatments markedly reduced macrophage M2 polarization in
   PTEN-deficient GBM models. We found that Gal-9 inhibition or anti–Tim-3
   antibody treatment impaired GBM growth by reducing macrophage M2
   polarization in vivo, while no direct impact on glioma cell
   proliferation was observed in vitro. We also established that
   PTEN-null–dependent Gal-9–driven up-regulation of macrophage M2
   polarization supports GBM growth in animal models.

   To explore the link between PTEN and Gal-9, we used several inhibitors
   to block multiple downstream signaling pathways activated by PTEN
   deficiency in glioma cells, including the AKT, SRC, YAP1, and NOTCH1
   pathways ([149]6, [150]23). We demonstrated that the AKT pathway played
   a key role in Gal-9 up-regulation. We then evaluated the molecules
   downstream of the AKT pathway that up-regulated Gal-9 expression and
   found that GSK3β, but not mTOR, participated in the up-regulation of
   Gal-9 in PTEN-KO glioma cells. Using biological databases coupled with
   an in vitro qRT-PCR–based validation method, we identified IRF1 that
   regulates LGALS-9 gene expression, likely through binding a specific
   promoter region. IRF1, a mediator of GBM resistance to bevacizumab
   ([151]24), was shown to regulate PD-L1 expression in cancer cells
   ([152]25). We found that IRF1 strongly correlates with the mesenchymal
   subtype GBM, similar to Gal-9/Tim-3. In addition to these findings,
   AKT-GSK3β-IRF–1driven mechanisms of Gal-9 up-regulation support the
   idea that genetic PTEN deficiency in glioma promotes macrophage M2
   polarization by regulating specific gene expression profiles.

   Although many previous studies ([153]6) used immunocompromised mouse
   models [BALB/c-nu/nu and nonobese diabetic–severe combined
   immunodeficient (Nod-SCID) mice] to examine the function of
   macrophages, such models are not yet realistic enough to mimic the
   macrophage-glioma interactions in humans in vivo. Therefore, we
   additionally used immunocompetent C57BL/6J mice and Wistar rats to
   establish orthotopic glioma (GL261-KO and C6) models, and treated the
   models with Gal-9/Tim-3 blockers. The results in both kinds of animal
   models verified that Gal-9/Tim-3 blockers were effective in impairing
   PTEN-null glioma progression.

   Growing evidence has shown that TAMs can promote glioma cell survival
   and angiogenesis ([154]8, [155]12). VEGFA secreted from macrophages
   plays a critical role in triggering angiogenesis ([156]26), and it is
   significantly up-regulated in macrophages isolated from GBM tissues
   ([157]27). Recently, a study showed that genetic ablation of
   macrophages delayed tumor angiogenesis and that VEGFA alone was
   sufficient to promote angiogenesis and compensate for macrophage
   deficiency ([158]28). Another study highlighted that knockout of VEGFA
   expression in myeloid cells attenuated glioma progression in
   experimental glioma models ([159]12). Here, we demonstrated that
   M2-polarized macrophages promoted glioma angiogenesis through VEGFA
   secretion. Unfortunately, antiangiogenic therapy of GBM with
   bevacizumab may accelerate glioma cell invasion and activate
   alternative angiogenic pathways ([160]29). In our study, analysis of
   scRNA-seq data of GBM tissues reveals that Gal-9 ligands activate Tim-3
   receptors on macrophages followed by activation of M2
   polarization–related TFs and up-regulation of M2 markers and
   angiogenesis-related genes in macrophages. Our results suggest that
   Gal-9/Tim-3 signaling–mediated glioma-macrophage interactions may
   represent as a novel target for glioma antivascularization. Here, the
   identification of VEGFA as an abundant TAM-secreted factor, along with
   α-lactose capable of decreasing VEGFA secretion, encourages the
   potential application of α-lactose in antivascularization treatment for
   GBM patients. Thus, our work reinforced the importance of TAMs in GBM
   angiogenesis and identified α-lactose as a potential and specific
   intervention targeting glioma angiogenesis.

   Given the known immunosuppressive and angiogenic role of TAMs, it is
   urgent to investigate whether GBM patients can benefit from treatments
   that decrease the number of M2 macrophages. Although CSF1R inhibition
   with BLZ945 alters macrophage polarization and blocks glioma
   progression ([161]15), the antitumor responses are transient and tumors
   recur in >50% of GBM mouse models ([162]16). A recent study reported
   that PI3K pathway activity was elevated in recurrent GBM, and combining
   PI3K blockade with CSF1R inhibition in these recurrent tumors
   significantly prolonged overall survival ([163]16). A clinical trial
   showed that the CSF1R inhibitor PLX3397 was not effective in patients
   with recurrent GBM ([164]17). Remarkably, PLX3397 extended the
   progression-free survival of two GBM patients (2/37), who presented
   with the mesenchymal subtype GBM with common gene alterations, such as
   PTEN deficiency ([165]17). Additional studies are ongoing, testing
   potential biomarkers to identify glioma patients with greater
   likelihood of response after macrophage-targeted therapy.
   Correspondingly, we found that Gal-9/Tim-3 signaling strongly
   correlated with mesenchymal subtype GBM, and predicted poorer clinical
   outcomes for glioma patients. Furthermore, we found that PTEN
   deficiency leads to up-regulated Gal-9 secretion by GBM cells via
   activation of the PI3K-AKT pathway, suggesting a novel mechanism of GBM
   resistance to macrophage-targeted therapy, such as CSF1R inhibitor
   treatment. We highlighted that α-lactose, a Gal-9 inhibitor, can impair
   glioma progression by reducing macrophage M2 polarization. The finding
   that Gal-9/Tim-3 signaling blockade increased survival in
   PTEN-deficient GBM models supports the importance of clinical testing
   of α-lactose specifically in PTEN-deficient GBM patients.

   Our in vivo data showed that blocking Gal-9/Tim-3 could significantly
   inhibit macrophage M2 polarization and angiogenesis, resulting in
   retardation of glioma growth. These results have significant
   therapeutic implications that targeting Gal-9/Tim-3 may be exploited as
   a precision immunotherapy specifically for PTEN-null gliomas.
   Furthermore, combining Gal-9/Tim-3 inhibitors with other existing
   targeted therapies (e.g., anti-VEGF antibody and anti–PD-1/PD-L1) is
   anticipated to produce a synergistic therapeutic effect for glioma. In
   future studies, it is necessary to conduct the clinical trials to
   confirm the efficacy of Gal-9/Tim-3 inhibitors in glioma.

   In conclusion, compared with PTEN-normal glioma cells, PTEN-deficient
   glioma cells recruited more macrophages and induced their polarization
   toward the M2 phenotype by abundantly secreting Gal-9 and activating
   Tim-3 on macrophages. Gal-9/Tim-3 blockade can thus impair
   glioma-macrophage interaction and attenuate angiogenesis to inhibit
   tumor growth.

MATERIALS AND METHODS

Human glioma samples

   A tissue array (240 glioma samples) containing 12, 85, 76, and 67 cases
   of grade I, II, III, and IV glioma was obtained from patients who
   received surgery at Sun Yat-sen University Cancer Center (SYSUCC)
   between 2001 and 2016 with written informed consents. All samples were
   pathologically diagnosed by experienced pathologists following the
   World Health Organization (WHO) 2016 Classification of Gliomas. The
   three human GBM tissues were collected for 10× Genomics scRNA-seq, and
   the PTEN gene deletion status was confirmed by fluorescence in situ
   hybridization performed in the Department of Molecular Diagnosis of
   SYSUCC (fig. S7). We also used Sanger sequencing to ascertain the PTEN
   gene mutation status of other human GBM tissues collected for primary
   culture. This study was approved by the Ethics Committee of SYSUCC (no.
   GZR2013-052).

Reagents and antibodies

   Information on all reagents and antibodies is shown in table S3.

Mice and xenograft tumor models

   All animal experiments were performed with the approval of the
   institutional ethics committee of SYSUCC (no. L102042020120S).
   Four-week-old Nod-SCID female mice were purchased from SPF
   Biotechnology Co. Ltd. (Beijing, China). Female BALB/c-nu/nu mice and
   C57BL/6 mice were purchased from Guangdong Medical Laboratory Animal
   Center (Guangzhou, China). Female Wistar rats were purchased from
   Zhejiang Vital River Laboratory Animal Technology Co. Ltd. (Zhejiang,
   China). The mice were inoculated with 1 × 10^6 cells (100 μl) in the
   right flank to establish subcutaneous xenografts, and the tumor weight
   was determined immediately after tumor removal at the end of the
   observation period. Intracranial tumor models were established as
   previously described ([166]30). A total of 1 × 10^5 cells were injected
   intracranially in 5 μl of 1× phosphate-buffered saline (PBS) into the
   mice, and tumor growth was monitored by bioluminescence imaging. After
   1 week of recovery, the glioma animal models were assigned to treatment
   groups. Mice received tail intravenous injection with 100 μg of
   purified anti-mouse Tim-3 antibody once a day for seven consecutive
   days. In addition, α-lactose (10 mg) was intraperitoneally injected
   into each mouse once a day for 14 consecutive days. Mice with moribund
   appearance were sacrificed for collecting brain samples. The samples
   were processed for embedding in paraffin blocks.

Cell culture

   The culture conditions and PTEN gene mutation status of all GBM cell
   lines used in this study are listed in table S1. Consistent with
   previous studies ([167]31, [168]32), our immunoblot results showed that
   C6 rat glioma cells did not express PTEN protein (fig. S16). THP-1
   macrophages, BMDMs, and human monocytes were cultured in RPMI 1640
   supplemented with 10% FBS. BMDMs from C57BL/6 mice were cultured as
   previously described ([169]33). Before transwell assays,
   differentiation was induced in THP-1 cells by phorbol 12-myristate
   13-acetate (PMA) (100 nM) for 72 hours, and differentiation was induced
   in BMDMs by CSF1 (10 ng/ml) for 6 days. Human monocytes were treated
   with CSF1 (10 ng/ml) for 3 days before use in in vitro experiments.
   After culturing glioma cells in FBS-free medium for 24 hours,
   supernatants were collected as the glioma cell–derived CM and then
   filtered with a polyethersulfone (PES) 0.22-μm membrane for use for in
   vitro experiments.

DNA extraction and DNA sequencing

   Genomic DNA was extracted from fresh human GBM tissue samples or cell
   lines using the TIANamp Genomic DNA Kit in accordance with the
   manufacturer’s instructions. DNA samples were sent to Guangzhou Tianyi
   Huiyuan Company for amplification and Sanger sequencing of the PTEN
   gene. PCR amplification of exons 1 to 9 of the PTEN gene was performed
   in a reaction volume of 30 μl containing specific forward and reverse
   primers (table S3). The PCR cycling conditions were as follows: initial
   denaturation at 94°C for 2 min, followed by 35 cycles of denaturation
   at 94°C for 10 s, annealing at 60°C for 30 s, and extension at 72°C for
   30 s, with a final extension at 72°C for 5 min. The PCR sequencing
   primers are listed in table S3. Regions of introns before and after the
   exons were also sequenced.

Migration assay

   Cells (1 × 10^5 THP-1 cells and 1 × 10^4 BMDMs) were suspended in
   serum-free medium and seeded into 24-well inserts (8.0 μm). GCM was
   added to the receiver wells. Cells were cultured for 24 hours, and the
   migrated macrophages were fixed with 10% formaldehyde, stained with
   crystal violet, and counted under a microscope.

Immunoblotting

   Cell lysates were subjected to SDS–polyacrylamide gel electrophoresis
   and transferred to polyvinylidene difluoride membranes. Blots were
   incubated with primary antibodies at 4°C overnight. After three washes
   with 1× Tris-buffered saline containing Tween 20 (TBST), the membrane
   was incubated with secondary antibody for 1 hour at room temperature.
   Enhanced chemiluminescence was used to detect protein signals. Band
   densities were normalized against the loading control β-actin,
   β-tubulin, lamin B1, or glyceraldehyde-3-phosphate dehydrogenase
   (GAPDH) using ImageJ software.

Flow cytometry

   Macrophages were pretreated with anti-mouse or human Tim-3 antibody, or
   α-lactose at the indicated concentrations, diluted with GCM for 72
   hours. Cells were collected and resuspended in 100 μl of PBS (1 × 10^7
   cells/ml). Fluorescence-conjugated antibodies were added, and the
   samples were incubated for 30 min at room temperature. After washing
   with 1× PBS, cells were analyzed by flow cytometry to detect CD68,
   CD86, CD163, and CD206 expression levels.

Enzyme-linked immunosorbent assay

   ELISA kits (MM-51152HI) were used to detect Gal-9 levels in CM
   collected from 23 glioma cell lines, including 13 PTEN-WT cell lines
   (UWR7, MGR2, SHG44, GL261, T98, SKMG-1, LN229, SKMG-4, SYU596, SYU488,
   SYU214, LN308, and SYU687), 8 PTEN-null cell lines (U118, SYU354,
   SF295, U373, MGR3, U87, U138, and U251), and 2 PTEN-mutated cell lines
   (U343 and A172). Glioma cells (1 × 10^7) were cultured in FBS-free
   medium for 24 hours; cell-free supernatants were collected for protein
   quantification with a bicinchoninic acid (BCA) protein assay kit,
   followed by evaluation using the ELISA kit.

Cytokine antibody microarrays

   The Quantibody Human Cytokine Antibody Array 440 was used to perform
   comprehensive proteomic analysis of the LN229-KO CM and LN229-WT CM.
   The cytokine profiles of PTEN-deficient and PTEN-normal GBM cells
   (PTEN-Del or PTEN-OE U87 cells; PTEN-WT or PTEN-KO SKMG-1 cells) were
   detected using RayBio C-Series Human T Cell Response Array 1. The
   cytokine profiles of BMDMs treated with GCM, GCM + α-lactose, or GCM +
   anti-mouse Tim-3 antibody were detected using Quantibody Mouse TH17
   Array 1. The detailed experiment protocols are available at
   [170]www.RayBiotech.com.

Single-cell RNA sequencing

Sample preparation

   GBM tissues from three patients were collected and mechanistically
   minced into approximately 1- to 3-mm^3 pieces. Samples were
   enzymatically digested in Dulbecco’s modified Eagle medium containing
   10% FBS, collagenase II (0.8 mg/ml; Sigma-Aldrich, catalog no. C6885),
   collagenase IV (1.5 mg/ml; Sigma-Aldrich, catalog no. C5138), 3 mM
   CaCl[2], deoxyribonuclease I (1 mg/ml; Sigma-Aldrich, catalog no.
   DN25), and 20% dispase buffer (Corning, catalog no. 354235) on a rotor
   at 37°C for approximately 30 min. The cell suspension was filtered
   through 70- and 40-μm cell strainer and then centrifuged for 4 min at
   400g. Cell pellets were resuspended in red blood cell lysis buffer
   (Boster, AR1118) and incubated on ice for 5 min. After centrifugation
   at 300g for 5 min, cell pellets were washed and resuspended in 1× PBST
   with 0.04% bovine serum albumin.

scRNA-seq library preparation and sequencing

   The single-cell suspension (15 μl, ~900,000 cells/ml) was loaded into
   one channel of the Chromium Single-Cell G Chip (10× Genomics, 1000120),
   aiming to capture 8000 to 9000 cells. The Chromium Single-Cell 3′
   Library & Gel Bead Kit v3.1 (10× Genomics, 1000121) was used for
   single-cell bar coding, complementary DNA (cDNA) synthesis, and library
   preparation, following the Single-Cell 3′ Reagent Kits User Guide
   (version 3.1). Libraries were then subjected to 150–base pair (bp)
   paired-end sequencing on an Illumina NovaSeq 6000 platform.

scRNA-seq data analysis

   We used Seurat v4.0.3 ([171]34) with default parameters to perform
   quality control, sctransform (SCT) normalization, integration, Uniform
   Manifold Approximation and Projection (UMAP) dimensionality reduction,
   and visualization for the scRNA-seq data. The “FindClusters” function
   was used for cell clustering (resolution parameter set to 0.1). The
   cell types were annotated on the basis of the expression of known
   marker genes ([172]35, [173]36). Specifically, the following marker
   genes (fig. S8) were used for cell type annotation: CD14, CD68, CSF1R,
   and FCGR3A for macrophages; SOX2, OLIG2, and ASCL1 for malignant cells;
   CD14, CD68, S100A8, and S100A9 for neutrophils; CD3D and CD3E for T
   cells; FBLN1 and DCN for fibroblasts; MAG and MOG for oligodendrocytes;
   and CLDN5 and VWF for endothelial cells. The annotation of malignant
   cells was verified by comparing the copy number patterns with
   nonmalignant cells ([174]37) using InferCNV v.1.6.0 with default
   parameters.

Multilayer network approach for cell-cell communication inference

   We developed an scRNA-seq data-driven multilayer network analysis
   approach to infer intercellular communications and intracellular gene
   regulations on the basis of our previous method scMLnet ([175]38,
   [176]39). The cell type–specific up-regulated genes were selected using
   the “FindAllMarkers” function in Seurat with default parameters. We
   collected prior information of ligand-receptor (LRDB) and TF-target
   gene pairs (TFTGDB) from Omnipath database ([177]40), a
   state-of-the-art knowledge base of intra- and intercellular signaling
   information. To construct receptor-TF pairs, we first extracted the
   protein-protein interaction from Omnipath as an undirected weighted
   graph G. The receptor-TF pairs were sorted according to the Dijkstra’s
   distance between receptors and TFs in the graph G, and the top 20% were
   used as confident receptor-TF pairs (RTFDB). The above LRDB, TFTGDB,
   and RTFDB were used as prior information for the following multilayer
   network construction.

   We leveraged scRNA-seq expression data to construct a cell
   type–specific multilayer network. For a pair of sender (cell type A)
   and receiver (cell type B), we ranked LR pairs on the basis of the
   product of mean expression of the corresponding ligand and receptor,
   and the top 20% of LR pairs were assumed as potential activated LR
   pairs. Fisher’s exact test was used to verify the activation of the TFs
   on the basis of the expression of the target genes (cell type
   B–specific up-regulated genes), which were then used to verify the
   activation of receptors. The filtering criteria were set as P value of
   <0.05. At last, the activated LR pairs, TFs, and the corresponding
   target genes were coupled together into a multilayer network. We
   performed the cell-cell communication analysis for each pair of cell
   types, and the number of the downstream target genes of activated LR
   pairs was used as the communication strength.

   To investigate functions of the cell-cell communication, we performed
   GO term (biological process) enrichment analysis for the Gal-9/Tim-3
   downstream target genes using clusterProfiler v3.18.1 ([178]41). In
   addition, we performed GSEA enrichment analysis to identify biological
   processes enriched by the ranked gene lists in macrophages on the basis
   of fold changes of gene expression in PTEN-null sample compared with
   PTEN-WT samples. Gene sets related to biological process classification
   in the GO database were downloaded from the MSigDB database for
   enrichment analysis.

Bulk RNA-seq data analysis

   For analysis of human GBM data, the gene expression and clinical and
   survival data of patients were downloaded from the CGGA database
   ([179]www.cgga.org.cn/). We analyzed the correlation between CD163 and
   VEGFA expression, between CD68 and LGALS9 (encoding Gal-9) or HAVCR2
   (encoding Tim-3) expression, and between IRF1 and LGALS9 expression as
   recorded in the mRNAseq_693 dataset. The expression of genes of
   interest (IRF1, LGALS9, and HAVCR2) in different GBM subtypes
   (mesenchymal, classical, neural, and proneural subtypes) was analyzed
   using the CGGA database.

   Using the GEPIA database ([180]http://gepia.cancer-pku.cn/), we
   confirmed the correlation between LGALS9 and IRF1 expression. We also
   analyzed LGALS9 expression in tumor and normal tissues in GBM and LGG
   patients.

   The gene expression profile and immune infiltration levels of GBM
   samples from The Cancer Genome Atlas (TCGA) database
   ([181]https://cancergenome.nih.gov/) was downloaded from UCSC Xena
   ([182]https://xenabrowser.net/datapages/) and TIMER2.0
   ([183]timer.cistrome.org). The gene expression data are presented as in
   log[2](x + 1) transformed RNA-Seq by Expectation-Maximization (RSEM)
   normalized count. GSEA of genes ranked by correlation with M2
   macrophage infiltration levels was conducted on the basis of VEGFA
   targets and VEGFA signaling gene sets
   ([184]www.gsea-msigdb.org/gsea/msigdb/genesets.jsp?collection=C6).

Survival analysis

   To assess the prognostic significance of Gal-9 and Tim-3 expression in
   glioma patients, we collected both gene expression data and the
   clinical information for glioma patients from three independent
   databases, including the CGGA database ([185]www.cgga.org.cn/), TCGA
   database ([186]https://cancergenome.nih.gov/), and Gene Expression
   Omnibus (GEO) database ([187]www.ncbi.nlm.nih.gov/geo/). The CGGA
   dataset, TCGA dataset, and [188]GSE16011 dataset included 400, 663, and
   264 samples, respectively. The patients in each dataset were stratified
   into two groups on the basis of Gal-9 expression, Tim-3 expression, or
   their product (Gal-9 × Tim-3), with an optimal cutoff value determined
   by using the receiver operating characteristic (ROC) method [i.e., on
   the basis of the highest value of (sensitivity + specificity)]
   ([189]42). Kaplan-Meier curves of overall survival for patients in the
   two groups were analyzed, and the statistical significance of the
   difference was assessed using the two-sided log-rank test.

Multiplex staining, multispectral imaging, and immunofluorescence

   To identify macrophage polarization in the TME of PTEN-deficient
   glioma, multiplex immunofluorescence staining of glioma tissue slices
   was performed using TSA Plus Fluorescence Kits combined with IHC
   (TSA-IHC) in accordance with the company protocols. The slices were
   sequentially incubated with primary antibodies (one by one) and
   horseradish peroxidase–conjugated secondary antibody (a mixture of
   anti-rabbit and anti-mouse). Slices were then incubated with TSA liquid
   reagent, which was respectively mixed with dyes (1:200), including
   PPD520 dye for staining Gal-9, PPD570 dye for staining Tim-3, PPD650
   dye for staining CD163, PPD690 dye for staining CD86, and
   4′,6-diamidino-2-phenylindole (DAPI) dye for staining nuclei. Before
   the final staining for nuclei, the slides were microwave heat–treated
   after each TSA operation. As for other operation details, the
   instructions of TSA-IHC kit could be followed. Slides were scanned at
   ×20 magnification for capturing multiplex IHC images or
   3,3’-diamino-benzidine (DAB)–stained IHC images using the Vectra
   Automated Quantitative Pathology Imaging System (Vectra 2.0.8;
   PerkinElmer). Five regions of interest (20×) were captured in each
   sample slice. First, the images were manually scored by a pathologist
   (S.X.), who served as a blinded reviewer. Tumor multispectral images
   were unmixed using a spectral library built from images of
   single-stained control tissues for each dye (PANOVUE) using the inForm
   software (inform 2.1.1; PerkinElmer). A selection of 10 representative
   multispectral images was used to train the inForm software. Then, all
   settings applied to the training images were saved as a project. All
   images were subsequently evaluated by the inForm analysis to confirm
   the manual scoring and calculate the accurate percentages of
   label-positive cells. In a sample slice, the positive percentages of
   five high-powered images (20×) were averaged and determined as the
   score (%) of the specific markers. A Gal-9–positive percentage of >25%
   was defined as Gal-9^High, and Tim-3-positive percentage of >5% was
   defined as Tim-3^High. (Gal-9 & Tim-3)^High was defined if the samples
   showed both Gal-9–positive percentage of >25% and Tim-3–positive
   percentage of >5%.

   Immunofluorescence in cells was performed using standard protocols. To
   determine the localization of Tim-3 and Gal-9 expression in
   macrophages, images were captured using a laser confocal microscope
   (Olympus).

qRT-PCR and RNA-seq

   For qRT-PCR, total RNA was extracted using the RNA Quick Purification
   Kit (ESscience, Shanghai, China) and reverse-transcribed to cDNA in a
   total volume of 20 μl, in accordance with the manufacturer’s
   instructions (catalog no. 11141ES60; Yeasen, Shanghai, China). All
   primer sequences for qRT-PCR are shown in table S3. qPCR (10 μl per
   well) was performed at 50°C (2 min) for Uracil-DNA Glycocasylase (UDG)
   activation and 95°C (2 min) for Dual-Lock DNA polymerase, followed by
   40 cycles at 95°C (15 s) for denaturing, 60°C (15 s) for annealing, and
   72°C (1 min) for extension using the Bio-Rad CFX96 Real-Time PCR System
   (Bio-Rad Laboratories Inc., Hercules, CA, USA). PowerUp SYBR Green
   Master Mix (01000439, Thermo Fisher Scientific) was used for signal
   detection. The number of cycles required to generate a given threshold
   signal (Ct) was recorded, and fold changes were calculated using the
   cycle threshold ΔΔCt method. Each reaction was performed in triplicate.
   The expression of each gene was normalized to that of mouse or human
   GAPDH expression.

   For RNA-seq, RNA was isolated from cells using TRIzol reagent
   (BCCC8200; Sigma-Aldrich) in accordance with the manufacturer’s
   instructions. RNA-seq analysis was performed with RNA prepared from
   PTEN-WT and PTEN-KO LN229 cells (biological triplicates for the control
   and PTEN-KO LN229 cells) at LC-Bio (a subsidiary of LC Sciences).
   Differentially expressed genes between the control and PTEN-KO LN229
   cells were subjected to GSEA. The up-regulated genes encoding secreted
   proteins in PTEN-KO LN229 cells were analyzed on the basis of the raw
   transcriptome data, and the results were validated by qRT-PCR.

PTEN knockout using CRISPR and PTEN overexpression

   A pool of three single-guide RNA (sgRNA) plasmids targeting human PTEN
   (sc-400103) was purchased from Santa Cruz Biotechnology, and a pool of
   three sgRNA plasmids targeting mouse PTEN (MCP227985-CG04-3-10) was
   purchased from GeneCopoeia Inc. (Rockville, MD, USA). The plasmids were
   transiently transfected into cells (LN229, SKMG-1, and GL261 cells)
   using Lipofectamine 3000. After 72 hours, the cells were harvested and
   green fluorescent protein (GFP)–positive cells sorted by
   fluorescence-activated cell sorting (FACS) were inoculated into wells
   of 96-well plates within one GFP-positive cell per well. PTEN-deficient
   single clones were confirmed by immunoblot, and a clone with successful
   PTEN knockout (PTEN-KO) was chosen for subsequent experiments. We
   overexpressed WT PTEN in glioma cells using lentivirus purchased from
   Obio Technology Corp. Ltd. (Shanghai, China).

shRNA-mediated knockdown

   Three shRNA hairpins targeting the human LGALS9 gene and a negative
   control shRNA were purchased from GeneCopoeia (Guangzhou, China).
   Recombinant lentiviral particles were produced using 293T cells.
   Briefly, 5 μg of the shRNA plasmid and 2 μg of VSVG, PLP1, and PLP2
   lentivirus packing plasmids were transfected into 293T cells plated in
   10-cm dishes using Lipofectamine 3000. Viral supernatant was collected
   at 48 and 72 hours after transfection and filtered for subsequent
   experiments. Cells were infected with viral supernatant containing
   polybrene for 48 hours and selected by incubation with puromycin (2
   μg/ml) for 2 weeks. The expression of Gal-9 in glioma cells was
   determined by immunoblots and qRT-PCR.

siRNA interference and dual-luciferase reporter gene assay

   siRNAs against human IRF1 (si-h-IRF1_02/06/07) and scramble siRNAs were
   purchased from Ribobio (Guangzhou, China). siRNAs were transfected into
   cells using Lipofectamine 3000 in accordance with the manufacturer’s
   instructions.

   The luciferase plasmids (pEZX-FR01-LGALS9-0.5/1.5/2.0 kb) and the
   control plasmids were purchased from GeneCopoeia. Three luciferase
   plasmids contain the 500/1500/2000-bp fragments of the LGALS9 promoter
   in the pEZX-FR01 vector containing the Firefly and Renilla luciferase
   genes. PTEN-KO LN229 and PTEN-KO SKMG-1 cells were transfected with
   pEZX-FR01-LGALS9-0.5/1.5/2.0 kb. PTEN-deficient U87 cells and
   PTEN-rescued U87 cells were transfected with pEZX-FR01-LGALS9-0.5 kb.
   In other experiments, pEZX-FR01-LGALS9-0.5 kb was cotransfected with
   siIRF1 or control siRNAs into PTEN-KO LN229 cells. At 48 hours after
   transfection, the cells were washed twice with cold 1× PBS, lysed in
   the lysis buffer provided in the dual-luciferase reporter gene assay
   kit (Yeasen, 11402ES60), and assayed for luciferase activity using a
   Tecan luminometer (Tecan Spark TM10M) in accordance with the
   manufacturer’s protocol. All transfections were performed in
   triplicate. Data are presented as the ratio of Firefly to Renilla
   luciferase activity.

Statistical analysis

   All statistical analyses were performed with Student’s t test or
   one-way ANOVA. Analysis of the survival data was performed using the
   log-rank test (GraphPad Prism 7). Data are presented as means ± SD, and
   P < 0.05 was considered significant.

Acknowledgments