The expression data of SYK in 33 tumor tissues and 21 tumor cell lines from the UCSC Xena platform (https://xena.ucsc.edu/), the Genotype-Tissue Expression (GTEx) portal (https://gtexport.org/home/), and the Cancer Cell Line Encyclopedia (CCLE) database (https://portals.broadinstitute.org/ccle/about) were accessed. We integrated the mRNA expression of SYK from the GTEx database into the TCGA dataset, and the expression data were normalized by log2 (expression values + 1) conversion.

In addition, the mRNA expression and clinical data of patients with diffuse gliomas were downloaded from four public datasets, including the Chinese Glioma Genome Atlas (CGGA) dataset (n = 693) (http://www.cgga.org.cn), TCGA dataset (n = 703), GSE16011 database (n = 276), and the Rembrandt database (n = 471) (https://www.ncbi.nlm.nih.gov/geo). In the TCGA cohort, a total of 677 samples were screened from 703 cases by excluding repeated samples.

A total of 175 diffuse glioma tissues, 37 paired tumors, and peritumoral tissues, and 22 normal brain tissues from traumatic decompression patients, were collected from Xiangya Hospital, Central South University. Informed consent was obtained from all patients, and this study was ethically approved by the ethics committee of Xiangya Hospital, Central South University.

We extracted total RNA using TRIzol (Life Technologies) from fresh tissues and synthesized complementary DNA from total RNA using a Reverse Transcription Kit (Thermo Fisher Scientific). Real-time qPCR was performed with ChamQ SYBR qPCR Master Mix (Vazyme). The relative expression of different sets of genes was quantified relative to ACTB mRNA using the 2^–ΔΔCt method. The primer sequences for qPCR used were listed as follows: SYK forward primer 5′-GCC​ATC​GGT​TCA​GTT​CA-3′, SYK reverse primer 5′-CAA​TCC​CCG​AGT​AAG​CAT-3′; ACTB forward primer 5′-ACA​GAG​CCT​CGC​CTT​TGC​CGA​T-3′, ACTB reverse primer 5′- CTT​GCA​CAT​GCC​GGA​GCC​GTT-3′.

A Tissue Microarray (TMA) containing 78 diffuse gliomas and six normal samples from Xiangya Hospital, Central South University, was constructed. Immunohistochemistry (IHC) was performed using a primary antibody against PD-L1 (Cell Signaling Technology Pathways, United States), CD163 (Proteintech, China), and SYK (Proteintech, China). The immunostaining results were evaluated by two independent pathologists based on the intensity and percentage of membranous or nuclear reactivity, as reported in the previous literature (Zhou et al., 2016). In summary, the target protein was mainly located in the nucleus and cytoplasm, and positive staining showed brownish-yellow or tan colors. Four different fields were randomly selected under high magnification (200×), and the total number of cells and the number of positive cells were counted. The staining was scored according to the following criteria: 0 if no cells were stained, 2 if 0%–25%, 2 if 25%–50%, 3 if 50%–75%, and 4 if more than 75% of the cells were stained. We scored the staining intensity as 0 for negative, 1 for weak, 2 for moderate, and 3 for strong. The total score was obtained by multiplying the percentage score by the intensity score. Two independent pathologists who were blinded to the source of the slides examined and scored each sample.

Samples with more than 30 days of overall survival (OS), disease-specific survival (DSS), and progression-free interval (PFI) were enrolled in the four datasets. The correlations between SYK expression and patient prognosis were assessed by the Kaplan–Meier curves in the four cohorts. The log-rank test was performed to assess the significance of SYK expression and clinical outcomes using the survival package in R. Univariate and multivariate Cox regression analyses were performed to estimate the independent prognostic factors, including age, gender, grade, SYK expression, and IDH status, in the TCGA cohort. p < 0.05 was considered statistically significant.

Based on the median expression value of SYK in the four cohorts, all patients were divided into high-expression and low-expression groups. The differentially expressed genes between the two groups were identified using the limma package in R software (version 3.6.0), with cutoff criteria of adjusted p < 0.05 and absolute log₂ (fold change) > 1 (Ritchie et al., 2015) in the TCGA cohort. Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis and Gene Ontology (GO) analysis of differentially expressed genes were performed using the package clusterProfiler (Yu et al., 2012). The GO categories included cellular components (CC), biological processes (BP), and molecular function (MF). In addition, gene set enrichment analysis (GSEA) was performed to investigate the underlying signaling pathways of SYK with GSEA software (vision 3.0), which was downloaded from the Broad Institute (http://www.broadinstitute.org/gsea). We obtained normalized enrichment scores and the p value of each biological pathway by GSEA.

Based on the gene expression profile, we applied the ESTIMATE R package to evaluate the immune and stromal scores, which can reflect the abundance of stromal cells and immune cells (Yoshihara et al., 2013). The abundance of six types of immune cells was obtained from the Tumor Immune Estimation Resource (TIMER, https://cistrome.shinyapps.io/timer/) (Li et al., 2017). The ssGSEA method was applied to evaluate the relative levels of 35 types of immune and stromal cells (Hänzelmann et al., 2013). The gene list of each immune cell type was obtained from recent publications (Bindea et al., 2013; Charoentong et al., 2017; Jia et al., 2018). In addition, we used the MCP-counter method to determine the absolute abundance of endothelial cells and fibroblasts (Becht et al., 2016).

We downloaded the single-cell data expression matrix of the GSE84465 database from GEO (https://www.ncbi.nlm.nih.gov/geo) (Darmanis et al., 2017), and analyzed the data with the R package of Seurat (Butler et al., 2018). After normalizing the expression data, we used “RunPCA” to perform principal component analysis (PCA). Eighteen cell clusters were identified and visualized by tSNE. We also analyzed the distribution of SYK and 10 markers of B cells, CD8⁺ T cells, CD4⁺ T cells, M1 macrophages, M2 macrophages, neutrophils, and dendritic cells in the cell clusters.

Unpaired Student’s t-test was performed to evaluate SYK expression between normal and tumor tissues, different grades, and wild-type and mutant IDH or IDH1, whereas the Kruskal–Wallis test was used to compare different histological subtypes. Pearson’s or Spearman’s test was conducted to analyze the correlation between SYK expression and the expression levels of the other genes. All statistical analyses were performed with GraphPad Prism 7.0 or R software (www.r-project.org). p < 0.05, p < 0.01, p < 0.001, and p < 0.0001 were considered statistically significant.

First, we analyzed the expression level of SYK in 32 types of normal tissues from the GTEx dataset, which suggested SYK was heterogeneously expressed in different normal tissues, with lower expression of SYK in the liver and muscle tissues, and the higher expression of SYK in the spleen, blood, and thyroid tissues (Supplementary Figure S1A) (p < 0.05). We also found that SYK was expressed at relatively low levels in the brain tissues (Supplementary Figure S1A) (p < 0.05). Furthermore, we explored the expression level of SYK in 21 types of tumor cell lines from the CCLE database, which indicated SYK was universally expressed in all tumor cell lines (Supplementary Figure S1B). To further explore the differences in SYK expression between tumor and normal tissues, we obtained SYK expression data from 25 types of tumors and normal tissues from the TCGA database. We found SYK was significantly up-regulated in the 11 types of tumors, including GBM and LGG, and down-regulated in seven types of tumor tissues (KIRC, KIRP, and LUAD) relative to normal tissues (Supplementary Figure S1C) (p < 0.05). However, no difference in SYK expression was observed in the seven types of tumor tissues, including BRCA, KICH, LICH, and SKCM, relative to normal tissues (Supplementary Figure S1C) (p > 0.05). To expand the number of normal tissues, we integrated SYK mRNA expression data from the GTEx database into those with the TCGA database. SYK expression was significantly higher in tissues from 25 types of tumors and lower in two types of tumor tissues (Supplementary Figure S1D) (p < 0.05). In conclusion, our results indicated that SYK was heterogeneously expressed in normal tissues and universally upregulated in pancancers tissues.

At present, few studies have investigated the role of SYK in diffuse glioma, and its regulatory mechanism in diffuse glioma is not clear. Considering that SYK was significantly differentially expressed in the diffuse glioma (Supplementary Figures S1C, D) (p < 0.05), we further explored the correlation between SYK expression and important clinicopathological features, such as WHO grade, the mutation status of IDH or IDH1, and histology, in the four cohorts. We found SYK expression was dramatically higher in higher grade diffuse glioma than in lower grade glioma, in the TCGA cohort (p < 0.05, Figure 1A). Similar results were observed in the CGGA dataset, Rembrandt dataset, and GSE16011 dataset (p < 0.05, Figures 1B–D). In the TCGA dataset, diffuse glioma with wild-type IDH had higher SYK expression than those with mutant IDH, which was validated in the CGGA cohort and GSE16011 cohort (p < 0.05, Figures 1E–G). In addition, high-expression SYK was frequently enriched in the GBM in the four cohorts (p < 0.05, Figures 1H–K). To validate the expression pattern of SYK, we also collected 22 fresh normal and 175 tumor tissues from Xiangya Hospital, Central South University, and 37 paired tumor and peritumoral tissues. In this validation dataset, SYK was highly expressed in the glioma tissues, compared with normal tissues (Figure 2A, p < 0.05). We also similarly found SYK was up-regulated in glioma with WHO grade IV (Figure 2B, p < 0.05), and wild-type IDH glioma (Figure 2C, p < 0.05) by qPCR, which was consistent with the aforementioned results. In addition, SYK was significantly over-expressed in 37 pairing tumor tissues, compared with matching peritumoral tissues (Figure 2D, p < 0.05). We then analyzed the SYK expression in the tissue microarrays that contained 78 diffuse glioma samples and six normal samples, by immunohistochemistry (IHC) staining. We found diffuse glioma with a higher WHO grade, wild-type IDH1, and GBM exhibited higher SYK expression (p < 0.05, Figures 2F–H). Therefore, these observations indicated that higher-expression SYK accompanies a more malignant phenotype in diffuse glioma.

To explore the clinical significance of SYK expression in diffuse glioma, the Kaplan–Meier curves were performed to assess the association between SYK expression and diffuse gliomas’ OS, DSS, and PFI in the TCGA cohort. The results revealed that high SYK expression predicted an unfavorable outcome in the TCGA cohort (log-rank test p < 0.001, Figures 3A–C). The correlation between SYK expression and OS was further validated in the other three cohorts (log-rank test p < 0.001, Figures 3D–F). In addition, the univariate and multivariate Cox regression analyses uncovered that SYK was an unfavorable prognostic marker, independent of age, gender, grade, and the mutation status of IDH in the TCGA cohort (the univariate Cox regression analysis, 95% CI = 2.033–3.849; the multivariate Cox regression analysis: 95% CI = 1.058–2.148) (Figures 3G, H, p < 0.05). Overall, these results suggested that SYK could serve as an independent predictor of a poor outcome among patients with diffuse gliomas.

To explore the potential pathways of SYK, all tumor samples were classified into low- and high-level groups based on the median value of SYK expression in the TCGA cohort. The differentially expressed genes between the two groups, including 3,684 significantly up-regulated genes and 2,910 significantly down-regulated genes, were identified with cut-off criteria of adjusted p < 0.05, and absolute log₂ (fold change) > 1 in the TCGA cohort (Figure 4A). GO and KEGG enrichment analyses indicated that significantly differential genes were enriched in the immunologic processes, such as “leukocyte migration,” “T-cell activation,” “regulation of immune effector process,” and “humoral immune response” (Figure 4B). We also performed GSEA to further explore the underlying pathways of SYK. The GSEA results suggested diffuse glioma with high-level SYK group exhibited biological pathways that were involved in modulating the immune response, including “IL2/STAT5 signaling,” “inflammatory response,” “NF-kappaB signaling,” and “IL6/STAT3 signaling” in the TCGA cohort (Figures 4C–F). Then, the ESTIMATE algorithm was used to comprehensively evaluate the abundance of two important components in the complex tumor microenvironment (Yoshihara et al., 2013). We also found diffuse gliomas with high-expression SYK consistently exhibited higher immune and stromal scores, compared with those with low-expression SYK in the four cohorts (p < 0.05, Figures 4G–J), which indicated SYK might be involved in regulating the immune and stromal cells in the tumor microenvironment. Therefore, our findings indicate that SYK is involved in remodeling the glioma microenvironment to promote malignant progression.

Considering that SYK might regulate the tumor microenvironment by immunologic biological processes, the role of SYK in the tumor microenvironment needs to be further explored in diffuse glioma. First, the correlations between SYK expression and the infiltration levels of six types of immune cells were evaluated in the GBM and LGG using TIMER website tools. We found that SYK expression was positively correlated with infiltrating levels of B cells (LGG, r = 0.748; GBM, r = 0.157), CD4+ T cells (LGG, r = 0.310; GBM, r = 0.917), macrophages (LGG, r = 0.819; GBM, r = 0.213), neutrophils (LGG, r = 0.840; GBM, r = 0.349), and dendritic cells (LGG, r = 0.890; GBM, r = 0.495) in the diffuse glioma (p < 0.05, Figure 5). However, SYK expression had an adverse correlation with the infiltrating levels of CD8+ T cells between LGG and GBM (p < 0.05, Figure 5). We then detected the distribution of different types of cells in the diffuse glioma with high- and low-levels of SYK expression. The abundance of 35 immune cell types was evaluated by the ssGSEA method in the TCGA cohort (Figure 6A). Our results uncovered diffuse glioma with high-expression SYK displayed a higher abundance of most immunosuppressive cells, except for CD56dim NK cells, in the TCGA cohort (p < 0.05, Figure 6B). Similar results were observed in the CGGA, GSE16011, and Rembrandt cohorts (Figures 6C–E, p < 0.05). However, no difference in abundance in the Th2 cells was observed in diffuse glioma with high- and low-expression SYK in the CGGA, GSE16011, and Rembrandt cohorts (Figures 6C–E, p > 0.05). In addition, we did not observe a difference in neutrophils in the GSE16011 and Rembrandt cohorts (Figures 6D, E, p > 0.05). Finally, we used the MCP-counter method to assess the absolute abundance of endothelial cells and fibroblasts in the four cohorts. We observed diffuse gliomas with high SYK expression showed a significantly higher abundance of endothelial cells and fibroblasts than those with low SYK expression (Figures 6F–I, p < 0.05). We did not find a significant difference in endothelial cells in the TCGA cohort (Figures 6F–I, p > 0.05). In conclusion, aberrant expression of SYK might be involved in shaping the immunosuppressive microenvironment in diffuse gliomas.

We performed the single-cell sequencing analysis to identify 18 cell clusters (Figure 7A), and further analyzed the distribution of SYK and 10 markers of B cells, CD8⁺ T cells, CD4⁺ T cells, M1 macrophages, M2 macrophages, neutrophils, and dendritic cells in cell clusters, such as CD163, VSIG4, CD19, CD8A, CD8B, CD4, ITGAM, NRP1, ITGAX, and NOS2. Our results uncovered that SYK and the markers of M2 macrophages, such as CD163 and VSIG4, shared a similar expression pattern, which suggested aberrant-expression SYK might recruit more M2 macrophages infiltration (Figure 7B). Consistently, we also found that SYK expression had moderate and strong correlations with the mRNA expression levels of CD163 and VSIG4 in the four cohorts (Spearman rank test, R > 0.500, p < 0.05, Figures 7C–F). In glioma tissue, diffuse gliomas with high-expression SYK exhibited high-expression CD163 by IHC (Figure 7G). In addition, a positive correlation between the expression of SYK and CD163 was observed (R = 0.330, p < 0.05, Figure 7H). These results indicated aberrant-expression SYK was associated with M2 macrophage infiltration in diffuse glioma.

To explore the efficacy of SYK as a synergistic partner of immune checkpoints, we first analyzed the correlation between SYK expression and the expression of immune checkpoints and immunogenic cell death in 33 types of tumors. The results showed that the expression of SYK was positively related to the expression of immune checkpoint molecules and immunogenic cell death, such as CD80, CD40, PD-L1, PD1, and PD-L2, in multiple tumors, including LGG and GBM (p < 0.05, Supplementary Figure S2). TMB serves as an important indicator of the efficacy of immunotherapy (Strickler et al., 2021). Consistent with the immune checkpoints, the mRNA levels of SYK were positively associated with tumor mutation burden (TMB) in multiple tumors, including GBM and MESO (p < 0.05, Figure 8A). Subsequently, we further analyzed the correlation between SYK expression and that of 37 immune checkpoint molecules and the corresponding ligands in the diffuse gliomas. We found SYK expression was tightly associated with some of the checkpoint molecules, such as CD276 (CGGA, r = 0.376; GSE16011, r = 0.354; Rembrandt, r = 0.369; and TCGA, r = 0.378), CD40 (CGGA, r = 0.581; GSE16011, r = 0.460; Rembrandt, r = 0.431; and TCGA, r = 0.474), CD44 (CGGA, r = 0.508; GSE16011, r = 0.632; Rembrandt, r = 0.599; and TCGA, r = 0.498), CD80 (CGGA, r = 0.571; GSE16011, r = 0.205; Rembrandt, r = 0.233; and TCGA, r = 0.489), CD86 (CGGA, r = 0.775; GSE16011, r = 0.752; Rembrandt, r = 0.734; and TCGA, r = 0.767), LAIR1 (CGGA, r = 0.800; GSE16011, r = 0.781; Rembrandt, r = 0.786; and TCGA, r = 0.758), and PD-L2 (CGGA, r = 0.621; GSE16011, r = 0.296; Rembrandt, r = 0.169; and TCGA, r = 0.565) (Table 1, Pearson test, p < 0.05). Although no correlation was found between SYK expression and PD-L1 expression in the GSE16011 and Rembrandt cohorts (Table 1, Pearson test, p > 0.05), a moderate positive correlation was shown in the TCGA (r = 0.369) and CGGA cohorts (r = 0.461) (Table 1, Pearson test, p < 0.05). Paradoxically, the expression of SYK was positively correlated with PDCD1 in the CGGA (r = 0.108) and TCGA datasets (r = 0.424) but negatively correlated in GSE16011 (r = −0.292) and Rembrandt cohorts (r = −0.118) (Table 1, Pearson test, p < 0.05). In addition, we further explored the association between SYK expression and some chemokines and immunogenic cell death (ICD) modulator-related genes and found the expression of SYK was closely correlated with the expression of most chemokines and ICD modulator-related genes in diffuse glioma (Pearson test, p < 0.05, Figures 8B,C). Finally, we explored PD-L1 expression using glioma tissue microarrays as mentioned earlier (Figure 2). A positive correlation was observed between SYK expression and the expression of PD-L1 (Pearson test, r = 0.300, p < 0.05, Figures 8D,E). Our findings suggest that blocking SYK might have improved synergistic effects with immune checkpoint inhibitors.

To screen patients with diffuse glioma suitable for blocking SYK, the SYK-related subtypes need to be identified. We defined 1,411 genes that were correlated with SYK expression (absolute Pearson r > 0.3, p < 0.001) by Pearson correlation analysis in the TCGA and CGGA cohorts (Figure 9A) (Supplementary Table S1). Based on the profiles of 1411 SYK-related genes, we performed NMF consensus clustering to identify three subtypes (Sub1, Sub2, and Sub3) in the TCGA cohort, which was validated in the CGGA cohort (Figures 9B, C). Importantly, diffuse glioma patients with subtype Sub1 (median survival, 17.3 months) had significantly shorter OS in the TCGA cohort, compared with those with subtype Sub2 (median survival, 63.0 months; HR = 0.31, 95% CI = 0.22–0.42, log-rank test p < 0.001) and Sub3 (median survival, 136.3 months; HR = 0.10, 95% CI = 0.07–0.16, log-rank test p < 0.001) (Figure 9D). Furthermore, similar prognostic difference was validated in the CGGA cohort, with diffuse glioma with subtype Sub1 (Sub1, median survival, 16.9 months) showing a significantly shorter OS than those with Sub2 (median survival, 26.0 months; HR = 0.60, 95% CI = 0.43–0.84, log-rank test p < 0.001) and Sub3 (median survival, 83.7 months; HR = 0.33, 95% CI = 0.27–0.41, log-rank test p < 0.001) (Figure 9E).

Next, we explored the correlations between the SYK-associated subtype and protumor immune cells (CD56dim NK cells, immature DCs, MDSCs, neutrophils, plasmacytoid DCs, Tregs, Th2 cells, activated mast cells, and M2 macrophages). Significant differences were observed between Sub1 and the other two subtypes, with a higher abundance of seven immune cell populations (immature DCs, MDSCs, neutrophils, plasmacytoid DCs, Tregs, Th2 cells, and M2 macrophages) for Sub1 compared with Sub2 or Sub3 in the TCGA cohort and CGGA cohort (Figures 9F, G). We speculated diffuse glioma with Sub1 was accompanied by an immunosuppressive tumor microenvironment. However, glioma with Sub1 did not exhibit a higher abundance of activated mast cells and CD56dim NK cells than Sub2 and Sub3 (Figures 9F, G).

We further investigated the association between subtypes and the expression of 13 potentially targetable immune checkpoint genes, such as PD-L1, PDCD1, and PD-L2, and the results showed that Sub1 exhibited higher expression of 13 immune checkpoint genes than Sub2 and Sub3 in the TCGA cohort (p < 0.05, Figure 9H). A similar difference was observed, with Sub1 showing significantly higher expression of immune checkpoint genes than that for Sub2 and Sub3 in the CGGA cohort (p < 0.05, Figure 9I). In addition, a waterfall plot was used to show the top 15 most frequently mutated genes in each subtype (Figures 9J–L). The mutation rates of IDH1 and ATRX were significantly lower in Sub1 (IDH1,18% and ATRX, 15%) than Sub2 (IDH1, 68% and ATRX, 44%) and Sub3 (IDH1, 88% and ATRX, 33%), while the mutation rates of PTEN, TTN and EGFR were significantly higher in Sub1 (PTEN, 24%, TTN, 22% and EGFR, 18%) than in Sub2 (PTEN, 9%, TTN, <5% and EGFR, 11%) and Sub3 (PTEN, <4%, TTN, 6% and EGFR, <4%) (Figures 9J–L). In conclusion, targeting SYK alone or in combination with other immune checkpoint inhibitors is more likely to show a benefit in diffuse glioma with Sub1 by improving the immunosuppressive tumor microenvironment and thus the patient prognosis.