We enrolled 288 glioma patients (WHO grade II-IV) operated at the Sun Yat-sen University Cancer Center (Guangzhou, China) from January 2009 to January 2016. The median follow-up time was 54 months. Follow-up was last done in June 2019. The detailed clinical data are listed in Table S1 . There were 167 males and 121 females. The median age of all patients at initial diagnosis was 43 years (range 7-78years). According to MRI imaging, 278 cases of glioma were located on the supratentorial and 10 cases of glioma were located infratentorial. 264 out of 288 patients received postoperative adjuvant treatment (radiotherapy or chemotherapy). The median overall survival time of all patients was 27 months (range 0-110 months). This cohort included 112 cases of astrocytoma, 45 cases of oligodendroglial gliomas and 131 cases of glioblastoma (WHO IV). All samples were ethically approved for use based on informed consent.

The human glioma cell lines and a human microglia cell line were purchased from the American Type Culture Collection resource center. The human glioma cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS) and the human microglia cells were cultured in Minimum Essential Medium (MEM) with 10% FBS at 37 °C in a humidified incubator containing 5% CO2. Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll-Hypaque density gradient centrifugation (Solarbio, Beijing, China). CD8+ T cell were separated by positive selection from PBMCs with CD8 magnetic beads and cultured in RPMI-1640 medium supplemented with 10% human serum, 5% L-glutamine-penicillin-streptomycin solution (Sigma-Aldrich, USA), CD3/CD28 antibody (Biolegend, USA) (25ul/ml) and IL-2 (100IU/ml) in 24-well plates. After culturing for 24 hours, add the corresponding concentration of human CHI3L2 protein (Sino Biological, Beijing, China) to T cells and culture for 72 hours at 37 °C in a humidified incubator containing 5% CO2.

Immunohistochemistry was essentially performed as previously reported (27). These tissue specimens were incubated with CHI3L2 rabbit polyclonal antibody (#22164, SAB, Maryland, USA). Immunohistochemical evaluation was independently conducted by two pathologists blinded for patient characteristics and outcome, and CHI3L2 expression by tumor cells and macrophage cells was scored separately. The discrepancies were resolved by consensus under a microscope for multi-viewing. A semi-quantitative IHC scoring criterion was used to determine the CHI3L2 protein expression levels in tumor cells. The percentage of positive cells and staining intensity were assessed to improve accuracy. The percent positivity of staining cells range from 0 to 4: 0, none; 1, 1%-25%; 2, 26%-50%; 3, 51%-75%; 4, 76%-100%. The intensity of staining was graded from 0 to 3 (0, no staining; 1, weak; 2, moderate and 3, strong). Then, we obtained the final IHC score by multiplying the proportion score by the intensity score of staining. We chose 4.5, which was determined by the Youden index as an optimal cutoff point to separate low CHI3L2 expression (score of 0-4.5) from high CHI3L2 expression (score>4.5) in tumor cells. For the macrophage cells, we only count the number of CHI3L2 positive staining macrophages, regardless of the staining intensity. We designed 7.5 determined by Youden index as an optimal cutoff point to differentiate low expression (number≤7.5) from high expression (number>7.5) in macrophages.

The other antibody markers including PHH3, P53, Ki67, ATRX, CD163, CD4, CD8, and CD20 were also used by immunohistochemistry tests. We detected MGMT promoter methylated status, TERT promoters and IDH mutation status by Sanger sequencing. 1p and 19q deletion status was detected using fluorescent in situ hybridization (FISH). The detailed protocol and assessment standard were described as a previous study (28).

The CHI3L2 RNA-seq data were downloaded from http://www.cgga.org.cn/. We totally analyzed 601 TCGA RNA-seq cohort and 608 CGGA RNA-seq cohort of gliomas, ranging from WHO grade II to grade IV.

To identify the CHI3L2-related genes, the limma package of R software was used to screen out the differentially expressed genes (DEGs). The top 26 hub genes of the overlapping DEGs were built via the plug-in molecular complex detection and cytoHubba of Cytoscape. To explore the functions and pathways of CHI3L2-related genes, we performed GO and KEGG analyses on ClueGo and Metascape websites. The TIMER platform (https://cistrome.shinyapps.io/timer/) (29, 30) was performed to explore the association between CHI3L2 and marker sets of tumor immune infiltrates in GBM and LGG (31).

Total protein was extracted from seven human glioma cell lines and one human microglia cell line (HMC3). 30 ug of protein was loaded onto 10% SDSPAGE and electrophoretically transferred to PVDF membranes. After blocking, the membranes were incubated with primary antibody against CHI3L2 (1:1000 dilutions, rabbit polyclonal anti-CHI3L2, #22164, SAB, Maryland, USA). The membranes were then incubated with horseradish peroxidase-linked anti-rabbit antibody (at a 1:3000 dilution, Santa Cruz Biotechnology, Inc., Santa Cruz, Calif., USA). B-Actin was served as a loading control.

Apoptosis was examined by flow cytometric analysis. Cells were collected, washed with PBS, and incubated with annexin V-FITC and PI (BD Biosciences, San Jose, CA, USA) for 15 minutes. Then cell apoptosis was analyzed by flow cytometry (FACS Calibur, Becton Dickinson, Sparks, MD) according to the manufacturer’s instruction. FlowJo (Treestar, USA) software was used for the analysis of flow cytometry data. The results are expressed as mean ± SD of three independent experiments.

GraphPad Prism 8 and SPSS 22 software were performed for statistical analyses. The measurement data are represented as mean ± SD. The Chi-square test was conducted to explore the correlations between CHI3L2 levels and clinicopathological features. Kaplan-Meier analysis was conducted for the overall survival of glioma patients with the log-rank test. The Cox proportional hazards regression model was used for univariate and multivariate analyses to evaluate the independence of CHI3L2 in predicting prognosis. The association between CHI3L2 and marker genes of immune infiltrating cells was assessed by Spearman's correlation coefficients. P < 0.05 was regarded as statistically significant.

We detected CHI3L2 protein expression levels in histological sections from patients with different glioma grades by immunohistochemistry (IHC). Among the 288 glioma specimens inspected, we found CHI3L2 was mainly stained in tumor cells, as well as macrophage cells. Figures 1A–D showed different IHC staining intensity of CHI3L2 in glioma tissues. Figures 1E, F mainly showed CHI3L2+ macrophages in glioma tissues. The IHC score of CHI3L2 in tumor cells and density of CHI3L2+ macrophages in different glioma subgroups are shown in Figure 2 . We found the expression levels of CHI3L2 in tumor cells were upregulated with increasing WHO grade of gliomas ( Figure 2A ), but there was no significant difference in CHI3L2+ macrophage density between WHO II and WHO III gliomas ( Figure 2D ). The CHI3L2 expression levels of GBM were significantly increased compared with LGG (WHO II-III) (P<0.001) in tumor cells ( Figure 2B ) and macrophages ( Figure 2E ). A significant increase of CHI3L2 expression levels was found in IDH-wildtype gliomas compared with IDH-mutant gliomas (P<0.001) in tumor cells ( Figure 2C ) and macrophages ( Figure 2F ). We further analyze the CHI3L2 IHC score and density of CHI3L2+ macrophages in diffusely infiltrating glioma of new molecular classification, including IDH mutant without 1p/19q codeleted gliomas, IDH mutant with 1p/19q codeleted gliomas, and IDH wild-type gliomas ( Figures S1A, B ). We found the expression of CHI3L2 is not related to the status of 1p/19q codeleted in IDH mutant gliomas.

Based on the expression levels of CHI3L2 in tumor cells and macrophage cells, we evaluated the association between CHI3L2 staining and clinicopathological factors, as listed in Table 1 . In tumor cells, we found significant correlations between CHI3L2 expression and WHO grade (P<0.001), age (P=0.001), Ki67 (P<0.001), P53 (P=0.034), PHH3 (mitotic figures) (P<0.001), ATRX protein expression (P=0.026), IDH (P<0.001) and 1p/19q codeleted (P=0.002). In macrophage cells, CHI3L2 is strongly correlated with WHO grade (P<0.001), gender (P=0.008), age (P=0.006), Ki67 (P<0.001), PHH3 (mitotic figures) (P<0.001), and IDH status (P<0.001). However, CHI3L2 expression levels of glioma cells were not significantly related to gender, location, TERT promoter mutation status, and MGMT promoter methylated status. In macrophage cells, CHI3L2 expression has no correlation with location, P53, ATRX protein expression, MGMT promoter methylated status, TERT promoter mutation, and 1p/19q codeleted status.

To explore the prognostic significance of CHI3L2 in gliomas, we performed the Kaplan-Meier method and log-rank test. We found high CHI3L2 expression levels of tumor cells and macrophages significantly predicted worse overall survival in diffusely infiltrating glioma (DIG) ( Figures 3A, D ) and lower-grade glioma (LGG) patients ( Figures 3B, E ). However, there was no statistical significance difference in GBM in our cohort ( Figures 3C, F ). When considering the CHI3L2 expression of tumor cells and macrophages together, we found the overall survival time was higher in patients with dual-low CHI3L2 expression in TC and MC compared to those in patients with non-dual CHI3L2 expression and dual high expression in DIG ( Figure 3G ), but this difference is not statistically significant in LGG and GBM ( Figures 3H, I ). Similarly, we also analyze the effect of CHI3L2 on the prognosis in the new molecular classification of glioma. We found CHI3L2 expression in tumor cells is closely related to the prognosis of all new molecular classification of glioma, and high CHI3L2 expression in tumor cells, macrophages and TC + MC predicted poor outcome for IDH wild-type gliomas ( Figure S2 ). Furthermore, we found regardless of whether patients with glioma have methylation of the MGMT promoter or have received adjuvant therapy, high CHI3L2 indicates a poor prognosis for glioma ( Figure S3 ).

Additionally, to evaluate the independent risk factors for prognosis of glioma, we conducted the univariate analysis ( Table 2 ) and multivariate analysis ( Table 3 ). In univariate analysis, CHI3L2 expression in tumor cells, CHI3L2+ macrophage cells density, CHI3L2 expression in both tumor cells and macrophage cells (TC + MC), grade, age, location, adjuvant therapy, Ki67 index, PHH3 (mitotic figures), IDH, and 1p/19q codeleted status were shown to be prognostic variables for the prognosis of overall survival in glioma patients ( Table 2 ). Then we included the prognostic variables in the univariate analysis (P<0.05) into the multivariate analysis. We found CHI3L2 expression in tumor cells, location, Ki67, IDH, 1p/19q codeleted were independent prognostic factors in gliomas ( Table 3 ).

To further verify the results of our study, we collected a total of 601 glioma samples from the TCGA dataset and 608 glioma samples from the CGGA dataset to analyze the CHI3L2 mRNA expression. In the TCGA dataset, CHI3L2 mRNA levels were significantly increased in GBM (WHO IV) compared with WHO II, WHO III, and LGG patients ( Figures 4A, B ). CHI3L2 mRNA expression levels were significantly higher in IDH wild-type gliomas compared with IDH mutant gliomas ( Figure 4C ). Similar results were also obtained in the CGGA dataset ( Figures 4D–F ). We further analyze the CHI3L2 mRNA expression levels in new molecular classification of diffusely infiltrating glioma, including IDH mutant without 1p/19q codeleted gliomas, IDH mutant with 1p/19q codeleted gliomas, and IDH wild-type gliomas, in TCGA and CGGA database ( Figure S4A and Figure S4B ). We found CHI3L2 mRNA expression levels in IDH wild-type gliomas are higher than IDH mutant gliomas. Gliomas with IDH mutant and 1p/19q codeleted have higher CHI3L2 mRNA levels than gliomas with IDH mutant and non-1p/19q codeleted. Moreover, we performed Kaplan-Meier analysis to confirm whether CHI3L2 mRNA levels could predict poor prognosis of gliomas in datasets. As shown in Figure 5 , patients with high CHI3L2 mRNA levels correspond to shorter survival time in all glioma subgroups both in the TCGA ( Figures 5A–C ) and CGGA ( Figures 5D–F ) datasets. Similarly, we also verified the effect of CHI3L2 on the prognosis in the new molecular classification of glioma in database ( Figure S5 ). Except for gliomas with IDH mutant and non-1p/19q codeleted in the TCGA dataset, high levels of CHI3L2 mRNA in any other subgroup indicate a poor prognosis, whether in the TCGA or CGGA dataset.

The GBM and LGG RNA-seq data were from TCGA and CGGA datasets. Limma package in R software was conducted to screen out the differentially expressed genes (DEGs) with the cut-off criterion of adjusted P< 0.05 and |log2FC| > 1. We identified 1356 overlapping DEGs which were aberrantly expressed in TCGA and CGGA datasets ( Figure 6A ). The top 26 hub genes were screened via the plug-in molecular complex detection and cytoHubba of Cytoscape ( Figure 6B ). GO analysis was performed to show the overlapping DEGs were involved in several biological processes, including angiogenesis, immune, and inflammatory response ( Figure 6C ). The KEGG pathways enriched in several classic signaling pathways, such as cell adhesion molecules (CAMs) and PI3K-Akt signaling pathways ( Figure 6D ).

Infiltrating immune cells are important components of the tumor microenvironment and are frequently associated with tumor behavior and patient outcomes. Since GO analysis revealed that CHI3L2 was related to the immune response, we further explored the infiltration of immune cells in gliomas. To estimate the relevance of CHI3L2 and diverse immune cell markers, we used the TIMER platform to investigate correlations between CHI3L2 levels and markers of diverse immune cells, included monocytes, TAMs (tumor-associated macrophages), M1 and M2 macrophages, Tregs (regulatory T cells), exhausted T cells, CD8+ T cells, T cells (general), B cells and neutrophils in GBM and LGG ( Table 4 ). We found CHI3L2 was significantly associated with marker sets of monocytes, TAMs, and M2 macrophages in GBM and LGG. Particularly, we showed the scatter plots of association between CHI3L2 and the marker sets of monocytes, TAMs, M1 phenotype, and M2 phenotype in GBM and LGG ( Figures 7A–H ). We also found significant correlations between CHI3L2 and some markers of Treg and T cell exhaustion, such as TGFβ1, CTLA4, TIM-3, and GZMB. Since Treg and T cell exhaustion play an important role in tumor immune escape. We believe CHI3L2 may also play an immunomodulation role in gliomas. In addition, we used several clinical commonly immune cell markers, including CD163, CD4, CD8, CD20, to perform immunohistochemical test on glioma samples, and found that CHI3L2+ macrophages have a certain correlation with CD163+ M2 macrophages (r=0.547, p<0.001), CD4+ T cells (r=0.330, p<0.001), CD8+ T cells (r=0.389, p<0.001), and CD20+ B cells (r=0.237, p<0.001) in gliomas ( Figure S6 ).

The expression of CHI3L2 in glioblastoma cells (U251, U87, T98G, DBTRG, A172, LN229) and microglia cell (HMC3) has been verified by Western Blot ( Figure 8A ). Figure 8A shows that CHI3L2 is expressed in glioblastoma cell lines and a microglia cell line. It is strongly expressed in the glioblastoma cell lines U87, U251, LN229, A172 and the microglia cell line HMC3, while the expression in the glioblastoma cell lines T98G and DBTRG is weak. Figure 8B is the result of flow cytometry analysis. The left image is a representative sorting that lists the percentage of cells in each quadrant: bottom left-live cells; top left-mechanically damaged cells; bottom right-early apoptosis; top right- late apoptosis. The cellular apoptotic rate was a sum of early and late apoptotic rates. The proportion of apoptotic cells in the control group was 28.1%, the proportion of apoptotic cells in the 0.5ug/ml CHI3L2 group was 33.6%, and the proportion of apoptotic cells in the 2.5ug/ml CHI3L2 group was 35.9%. The right chart shows the percentage of apoptotic cells of three independent experiments. The result of flow cytometric analysis demonstrated that the apoptotic proportion of CD8+ T cells increases with increasing CHI3L2 concentration, indicating CHI3L2 may be related to immunosuppression.