3.1 TOX expression is decreased in maligant gliomas
Mining expression data from publicly available data-bases: TCGA, n = 674; CGGA, n = 1017, we evaluated the mRNA expression levels of TOX in WHO grade I-IV gliomas.
First, we analyzed the expression pattern of TOX across grades and subtypes in TCGA and CGGA datasets. We evaluated TOX levels in various common cancer types including gliomas (Fig. 1A). Compared to normal brain tissues, tumor samples demonstrated significantly up-regulated TOX expression, suggesting its adverse role in glioma development and progression. TOX was significantly elevated in low grade glioma (LGG) samples compared with GBM samples (Fig. 1C). Interestingly, TOX had the highest expression in WHO grade III samples in both the TCGA and CGGA datasets (Fig. 1B).
TOX was down-regulated in 1p/19q non-codeletion pan-glioma analysis, but up-regulated in the 1p/19q codeletion pan-glioma analysis in both TCGA and CGGA cohorts (Fig. 1D). Similarly, IDH mutation, indicating better clinical outcome, had tight association with a high expression level of TOX (Fig. 1E). Furthermore, in WHO grade II glioma samples, the IDH mutation status was significantly related to higher expression of TOX in the TCGA and CGGA cohorts (Fig. 1E). The ROC curve further suggested that TOX could be an valuable predictor for IDH mutation among pan-gliomas analysis, LGG cases, and GBM cases respectively (AUC value = 0.878, P < .001; value = 0.841, P < .001; value = 0.814, P < .001, respectively Fig. 1F). Notably, in LGG samples, IDH mutation together with 1p/19q codeletion is related to higher expression of TOX in both TCGA and CGGA cohorts (Fig. 1G). In addition, higher expression of TOX was related to methylated glioma in TCGA cohort (Fig. 1H). In CGGA cohort, females had relatively higher expression levels of TOX (Fig. 2A). The different expression level of TOX in glioma in regard to histology was shown in Fig. 2B.
3.2 Molucular characteristics of TOX in gliomas
The molecular categorization of human gliomas has four distinct sub-classes: mesenchymal (MES), classical (CL), neural (NE), and proneural (PN). MES and CL subtypes are related to more aggressive behavior of gliomas and more dismal clinical outcome of patients compared with PN or NE subtypes[29] [30]. Therefore, we subsequently analyzed the expression level of TOX among these four molecular subtypes on the basis of VERHAAK_2010 classification scheme [31].
In the TCGA dataset, higher TOX expression was seen in MES and CL subtypes of GBM compared to NE and PN subtypes, while the distinction was conspicuous in LGG samples and pan-glioma analysis (Fig. 2C). The ROC curve further indicated that TOX might serveas a predictor for CL and MES subtypes in pan-gliomas analysis, LGG, and GBM samples (AUC value = 0.883, P < 0.001; value = 0.860, P < 0.001; value = 0.695, P < 0.001, respectively Fig. 2E). Moreover, the highest TOX expression was seen in the PN molecular subtype (Fig. 2C).
We next evaluated the intra-tumour distribution of TOX in GBM samples. Based on the TCGA dataset, the analysis of RNA sequencing data revealed the high expression of TOX in cellular tumour, leading edge and infiltrating tumour (Fig. 2D). To further confirm the upregulation of TOX expression at the protein level, we downloaded the results of IHC staining for TOX from the The Human Protein Atlas (https://www.proteinatlas.org) (Fig. 2F). TOX has higher expression in LGG and GBM compared to normal brain tissue. The expression of TOX is also higher in LGG than GBM, which is consistent with our results.
3.3 TOX expression predicts better survival probability
We further investigated the prognostic value of TOX in human gliomas. Based on the calculated median values of TOX expression in gliomas, we generated Kaplan-Meier survival curves. In TCGA GBM dataset, TOXhigh patients exhibited significantly longer overall survival (OS), disease specific survival (DSS), and progressive free survival (PFS) compared with TOXlow patients (P < 0.05, respectively; Fig. 3A,3B,3C). In addition, in TCGA LGG datasets, TOXhigh patients exhibited significantly longer overall survival (OS), disease specific survival (DSS), and progressive free survival (PFS) compared with TOXlow patients (P < 0.001, respectively; Fig. 3D,3E,3F). This result was further confirmed in pan-glioma analysis (P < 0.001, respectively; Fig. 3G,3H,3I). In CGGA dataset, TOXhigh patients were associated with longer OS in pan-glioma, LGG, and GBM analyses (P < 0.001, P < 0.001 and P < 0.05 respectively; Fig. S1A, S1B, S1C). Furthermore, Cox regression analysis was performed to explore the clinical prognostic value of TOX in gliomas. In the Univariate analysis, TOX, together with high WHO Grade, age at diagnosis, 1p19q codeletion, and IDH mutations were significantly related to OS in both TCGA and CGGA databases (Table 1,2). In the multivariate analysis, TOX was also proved to be a valuable predictor in both cohorts. These results revealed that TOX might serve as a predictor for the better prognosis of glioma patients.
Table 1
Univariate and multivariate cox analyses in gliomas in CGGA.
Factor | CGGA RNA-seq set |
| Univariate | Multivariate |
| P | HR | 95%CI | P | HR | 95%CI |
TOX High vs. Low | < 0.001 | 2.34 | 1.95–2.82 | < 0.001 | 1.44 | 1.18–1.76 |
Age Increasing years | < 0.001 | 1.03 | 1.02–1.04 | 0.044 | 1.01 | 1.00-1.02 |
Gender Male vs. Female | 0.823 | 1.02 | 0.85–1.22 | 0.518 | 0.98 | 0.82–1.17 |
WHO Grade | | | | | | |
Grade III vs. II | < 0.001 | 3.01 | 2.26-4.00 | < 0.001 | 2.90 | 2.17–3.86 |
Grade IV vs. II | < 0.001 | 8.53 | 6.40-11.01 | < 0.001 | 5.22 | 3.88–7.02 |
1p19q status Codel vs. Non-codel | < 0.001 | 4.41 | 3.26–5.97 | < 0.001 | 2.56 | 1.85–3.55 |
IDH status Mutation vs. wild-type | < 0.001 | 3.12 | 2.59–3.75 | 0.023 | 1.23 | 0.99–1.54 |
Factor
|
TCGA RNA-seq set
|
|
Univariate
|
Multivariate
|
|
P
|
HR
|
95%CI
|
P
|
HR
|
95%CI
|
TOX
High vs. Low
|
< 0.001
|
4.32
|
3.24-5.77
|
0.044
|
1.47
|
1.01-2.14
|
Age
Increasing years
|
< 0.001
|
1.06
|
1.05-1.07
|
< 0.001
|
1.03
|
1.02-1.04
|
Gender
Male vs. Female
|
0.084
|
1.26
|
0.97-1.63
|
0.229
|
1.20
|
0.90-1.56
|
WHO Grade
|
|
|
|
|
|
|
Grade III
|
< 0.001
|
3.34
|
2.28-4.89
|
< 0.001
|
2.20
|
1.48-3.26
|
Grade IV
|
< 0.001
|
17.95
|
12.11-26.59
|
< 0.001
|
3.64
|
2.23-5.95
|
1p19q status
Codel vs. Non-codel
|
< 0.001
|
4.23
|
2.75-6.51
|
0.012
|
1.88
|
1.15-3.10
|
IDH status
Mutation vs. wild-type
|
< 0.001
|
8.92
|
6.76-11.75
|
< 0.001
|
2.27
|
1.47-3.51
|
Table 2. Univariate and multivariate cox analyses in gliomas in TCGA.
3.4 The association between TOX expression levels and distinct genomic alterations
We next performed somatic mutation analysis and copy number variation (CNV) using the TCGA dataset to determine whether TOX expression levels were associated with specific genomic characteristics. By comparing the TOXlow (n = 158) and the TOXhigh (n = 158) clusters (Fig. 4C), we obtained an overall CNV profile. The amplified chromosome 7 and the deleted chromosome 10, two most common genomic events in GBM, both were frequently associated with the TOXlow cluster (Fig. 4A). The genomic hallmark of oligodendroglioma, the deletion of 1p and 19q, was more frequently occurred in the TOXhigh cluster (Fig. 4A).
We next identified 43 and 61 genomic events enriched in either the TOXhigh or TOXlow group using GSITIC analysis (Fig. 4B). In TOXlow samples, oncogenic driver genes including PIK3C2B (1q32.1), PDGFRA (4q12), EGFR (7p11.2), and CDK4 (12q14.1) were frequently amplified genomic regions. Meanwhile, frequently deleted genomic regions included tumour suppressor genes such as PARK7 (1p36.23), CDKN2A (9p21.3), and PTEN (10q23.3). In TOXhigh samples, 8q23.3 and 12p32.32 were two significant amplified peaks, while significant deletion showed peaks in 2q37.3, 4q35.2, 9p21.3, 11p15.5, and 19q13.43. Notably, a 4q12 peak was detected in both TOXhigh and TOXlow samples. However, the G score in TOXhigh samples was obviously higher than that in TOXlow samples. Based on TOX expression levels, the somatic mutation profiles revealed that mutations in IDH1 (91%), CIC (28%), and ATRX (37%) were significantly enriched in GBM samples with high TOX expression (Fig. 4C). In addition, frequently observed mutations were EGFR (27%), IDH1 (20%), PTEN (18%), and MUC16 (16%) in gliomas with low TOX expression (n = 158; Fig. 4C).
3.5 TOX is involved in complicated immune processes
We further investigated the potential immune-related functions of TOX in glioma using GSVA analysis in TCGA dataset. In GBM alone, we found that TOX was positively associated with B cell activation, T cell receptor signaling pathway, B cell homeostasis, and T cell proliferation. In contrast, TOX had negative association with lymphocyte migration, natural killer cell activation, and lymphocyte chemotaxis. (Fig. 5B) In pan-glioma analysis, TOX had negative association with T cell migration, Negative T cell selection, Natural killer cell mediated immunity, Regulation of T cell cytokine production, Positive regulation of T cell apoptotic process, B cell mediated immunity, Lymphocyte migration, and Lymphocyte chemotaxis.(Fig. 5A) In LGG alone, TOX was negatively related to T cell migration, Lymphocyte migration, Regulation of T cell cytokine production, lymphocyte mediated immunity, and regulation of αβ T cell proliferation. Similar results were seen in CGGA dataset. (Fig. 5C)
A previous study has demonstrated that TOX is essential in the development of Innate Lymphoid Cells [32]. Consequently, we paid special attention to two pathways mentioned above: lymphocyte migration and lymphocyte chemotaxis. As the threshold was set as logFC > 2 and adjust P ≦ 0.01, a total number of 2778 differentially expressed genes (DEGs) were detected between high expression of TOX sample and low expression of TOX sample (Fig. 5D). As for lymphocyte migration, eight genes were found involved in both DEGs and lymphocyte migration gene sets. SAA1, CXCL11, CXCL10, CCL2, CCL20, CXCR3, and MYO1G were related with high expression of TOX, whereas RET was related with low TOX expression. As for lymphocyte chemotaxis, expression of TOX was negatively related to SAA1, CXCL11, CXCL10, CCL2, CCL20, CXCR3 (Fig. 5E,5F).
3.6 TOX is irrelevant to inflammatory activities
We examined the association between the molecules and various molecules related to inflammatory activity in both TCGA and CGGA datasets. TOX expression was negatively associated with inflammatory activity signatures including HCK, LCK, MHC-I, MHC-II, STAT1, and interferon metagenes, but positively associated with the IgG metagene in pan-glioma analysis, LGG alone and GBM alone (Fig. 6A,6B,6C; Fig. S1D, S1E, S1F, S1G).
These results indicated that TOX was not involved in signaling transduction of T cell activation, macrophage activation, or antigen presenting cells (APCs). However, TOX might have interaction with B lymphocytes in the process of immune-activation and subsequent glioma suppression.
3.7 The tight associated between TOX and immune cells in the tumour microenvironment
We further examined the significance of increased TOX in immune-related microenvironment in gliomas via performing GSVA analysis. We identified the immune cell types in the microenvironment of gliomas to see if they are influenced by TOX and to evaluate its presumed role in the interaction between gliomas and immune cells. We first investigated the relationship between TOX and 28-immune cell populations using cell type gene set variation analysis[33]. In both TCGA and CGGA cohorts, we found that TOX was positively associated with Eosinophil in pan-glioma analysis, whereas multiple immune cell types with infiltration charateristics, macrophages, monocytes, CD4 + TEM, CD8 + T effector memory cells (TEM), neutrophils, Myeloid-derived suppressor cells (MDSC), and natural killer (NK) cells were negatively associated with TOX in the pan-glioma analysis and in the LGG analysis (Fig. 7D, 7F; Fig. S3, S4). For GBM samples, DCs, MDSC, macrophages, mast cell, NK cells, CD8 + TEM, and CD4 + TEM etc. were found to be negatively associated with TOX (Fig. 7B; Fig. S3, S4). We further validated these results in a 24-immune cell lineage analysis, confirming the rejection of multiple immune cell types[34] in TOXhigh glioma samples. In 24-immune cell lineage analysis, Neutrophils, Eosinophils, Macrophages, NK cells, and DCs were negatively associated with TOX. TFH (follicular helper cells) and tumor growth delay (TGD) were positively associated with TOX in the pan-glioma analysis and the LGG group (Fig. 7C, 7E; Fig. S3, S4), while TFH and B cells were positively associated with TOX and Macrophages and DCs were negatively associated with TOX in GBM samples (Fig. 7A; Fig. S3, S4). Altogether, our data revealed that the high expression of TOX tend to reject the infiltration of immune cells in the microenvironment of gliomas.
3.8 TOX is synergistic with other immune checkpoint members
Given that the immune checkpoint molecules vitally regulate immune processes, we assessed the correlation between TOX and several crucial immune checkpoints in glioma samples. TOX was strongly correlated with CD276, IDO1, PDCD1LG2, and VTCN1 in pan-glioma analysis and GBM alone in both TCGA and CGGA cohorts (Fig. 8A,8B; Fig. S5), which the correlation was significantly better in LGG samples alone (Fig. 8C; Fig. S5). The analysis of TOX family showed favorable inter-relationship between TOX, TOX2, TOX3, and TOX4 in pan-glioma analysis, LGG samples, and GBM samples (Fig. 8D,8E,8F; Fig. S4).