Identification of 3 subpopulations of tumor-infiltrating immune cells for malignant transformation of low-grade glioma

Gliomas account for 60% of all primary and other central nervous system (CNS) tumor diagnoses, and make up ~ 80% of all malignant brain tumors [1]. The World Health Organization (WHO) classifies gliomas according to histology and molecular subtype, and grades them by the scale of I, II, III, IV. low-grade gliomas (LGG) typically range from grades I–II, while high-grade gliomas (HGG) are categorized as grades III–IV. Glioblastoma multiforme (GBM) is grade IV glioma subtype which often spontaneously appears in the CNS, but can also progress from LGG. GBM takes up half of CNS tumors, and is a fatal disease with no curable therapy [2]. Even with a comprehensive therapy, such as surgical resection, adjuvant radiotherapy, and alkylating agent temozolomide chemotherapy, patients who suffer from gliomas still have short median survival time, due to the aggressiveness of tumors, resistance to treatments, and recurrence over time [3]. In particular, patients with GBM approximately has a median survival of only 14–16 months [4]. In the past decade, studies on the anticancer immune responses for other tumors have promoted clinical advances in the limited success of conventional therapies. Meanwhile, the discovery of CNS lymphatic system has provided a new theoretical basis and opportunity for brain tumor immunotherapy [5].

Tumor-infiltrating immune cells (TIICs), whose function and composition subtly altered with the immune status of the host have been reported to be effectively targeted by drugs correlate with clinical outcome [6]. Melanoma and non-small-cell lung cancer are the two solid tumors in which immunotherapy has proved to be effective [7]. However, compared with these two tumors, glioma harbors a lower burden of somatic mutations and a more immunosuppressive tumor microenvironment [8]. Unique challenges should be overcome before immunotherapy applied to CNS. First, anatomically, the blood–brain barrier (BBB) restricts the entry of immune cells to the brain parenchyma. Also, the tumor cells themselves secrete a variety of immunosuppressive factors that influence macrophage polarization, dendritic cell (DC) maturation, regulatory T cell recruitment, inhibition of neutrophil and natural killer (NK) cell function. Previous studies have revealed that glioblastomas are heavily infiltrated with monocytes/microglia, although TIICs are relatively rare. Reports suggest that these cells account for 10–30% of viable cells within the tumor mass. They appear to be affected by tumors and have positive immunosuppressive effects. For example, Rodrigues et al. demonstrated that normal monocytes that come into contact with glioblastoma cells secrete multiple immunosuppressive factors (IF-10, TGF-β, B7-H1), have reduced phagocytic ability and induce apoptosis in activated T cells [9]. While preclinical data shows the success of immunotherapy for gliomas, the profiles of TIICs in glioma and their clinical value still remain to be explained.

Nevertheless, Immunohistochemistry and flow cytometry are the two most commonly used techniques that depend on a single marker for detecting TIICs in previous studies. Obviously, these approaches can be misleading and are not comprehensive as many marker proteins are expressed in different cell types. The “Cell type Identification By Estimating Relative Subsets Of RNA Transcripts” (CIBERSORT) employs deconvolution of bulk gene expression data and a sophisticate algorithm for in silico quantification of many immune cell types in heterogeneous samples as tumor stroma. Here, we used CIBERSORT, for the first time, to quantify the 22 TIICs subpopulations of immune response in glioma based on the patients’ gene expression profiling from TCGA and CGGA public databases in order to investigate its relationship between clinicals factors, with the final goal of developing new immunotherapeutic strategies.

For a long time, although it is well known that immune cells play an important role in tumor initiation and development, these insights have few influence on clinical practice [11, 12]. In addition, the role of genes that are abnormally expressed in tumor tissues in diagnosis and prognosis has also attracted widespread attention; however, few studies have focused on the differential distribution of immune cells between different components. In this paper, we firstly established an immune risk score model based on the fractions of three subpopulations of TIICs. Compared with the high-IRS group based on our model, the low-IRS group has a significantly better survival rate (p < 0.001). This finding suggests that our IRS model can better predict progression of glioma, especially in the MT from LGG to GBM. Validation group, IHC and functional enrichment analyses further illustrate the validity of the model. This study opens a door for a better understanding of new diagnosis strategy from the perspective of TIICs. We acknowledge that there exist limitations in this research, particularly no precise analysis of the effect of single TIICs. Besides, studies on TIICs in initiation of glioma weren’t carried on due to lack of sequencing samples from normal people in these public databases. Therefore, further studies are urgently needed to analysis single TIISs and whether it is possible to detect the real-time progression of tumor through the state of immune cells in the circulatory system.

Gliomas are tumors of CNS, originating from transformed neural stem or progenitor glial cells [13]. On the basis of histopathological characteristics WHO classified gliomas into groups: low-grade gliomas (LGG, grades I and II) are well differentiated, slow-growing tumors, whereas high-grade gliomas (HGG, grades III and IV) are less differentiated or anaplastic, and strongly infiltrate brain parenchyma [14]. Glioblastoma (GBM) is categorized as the most malignant type (grade IV). It accounts for 50% of CNS tumors and is a deadly disease without curable therapy. Despite aggressive treatments, such as extensive resection combined with radiation and/or chemotherapy, patients with GBM eventually die of their disease [4]. In another aspect, patients with LGG may survive for many years, but after transformed to GBM, survival rates rapidly decline [15, 16]. A population-based study showed that the mean period of malignant transformation from LGG to GBM was 5.3 years and for anaplastic astrocytoma to GBM was 1.4 years [17]. Most of the predictive models established in previous studies on glioma development and malignant transformation were based on differential expressed genes, but they neglected that immune cells may also play an important role in tumorigenesis. Due to technical limitations, previous researches were limiting to a narrow insight of tumor-infiltrating cells. Immunohistochemistry and flow cytometry which depend on a single surface marker were used to evaluated TIICs. Apparently, these techniques may have misidentified other cell with the same surface markers as TIICs and are subjectively affected by observers. Thus, in the current study, we employed a silicon analysis, known as CIBERSORT, to infer the proportions of 22 immune cell subpopulations from glioma transcriptomes. CIBERSORT is a deconvolution algorithm for charactering TIICs composition of complex tissues by analyzing 547 gene expression, introduced by Newman etc. in 2015. They firstly employed a novel application of liner support vector regression to deconvolve the tissue composition. In order to assess the feasibility of TIICs deconvolution from bulk tumors, they then designed and validated a TIICs gene signature matrix, termed LM22. By using LM22 to deconvolve 3061 human transcriptomes, they therefore proved CIBERSORT has great specificity and sensitivity [11]. As an emerging technology, CIBERSORT have already conducted in breast cancer [18], lung cancer [19], colon cancer [6] and so on, all these studies demonstrated the effectiveness and accuracy of this tool when analyzing TIICs.

Univariate and multivariate analyses indicated TFH cells, activated NK cells and M0 macrophages as independent predictors. Then, based on their correlation coefficients, we firstly constructed such an IRS model in glioma. Among these correlation coefficients, or the degree to which the cell distribution correlated with tumor progression, the coefficient of activated NK cells is negative, while the coefficients of the other two TIICs are positive. This is consistent with our previous analyses between TIICs and tumor grade. Hence, we have adequate reason to believe that this model can predict MT between LGG and GBM well.

Immune system can be functionally divided into innate immunity and adaptive immunity, where adaptive immunity is antigen-specific. It mainly consists of B cell-mediated humoral immunity and cytotoxic T cell-mediated cellular immune responses, and both these two adaptive immunity processes require signals from CD4 T cells [9]. In one aspect, some CD4 T cell subpopulations such as Th1 cells can exert anti-tumor immunity by overcoming the tolerance of autoantibodies expressed by tumors, and these effectors T cells are advantageous for tumor immunotherapy [20]. However, other subsets of CD4 T cells, particularly regulatory T cells and TFH cells, inhibit tumor immunity, thereby promoting cancer growth [21‐23]. In our study, although there was no significant difference in the composition of TFH cells between LGG group and HGG group, but in Fig. 2b we can see that its level in the GBM group is higher than that of the lower grade gliomas. At present, there is no research on the role of TFH in the immune microenvironment of glioma, which is the problem we need to think about and solve next.

Unlike T cells, NK cells play a unique role in innate and adaptive immune responses without the involvement of major histocompatibility complex (MHC) antigens or antibodies [24], and monitor status of intracellular bacteria, viruses-infected cells and transformed cells. Activated NK cells are one of two types of lymphokine-activated killer (LAK) cells. When stimulated by IL-2, they become activated against tumor cells. Although no randomized controlled trails of immunotherapy with HGG by LAK has been performed to date, one study showed patients treated with LAK cells had longer survivals than control groups [25]. Due to the difficulties in producing sufficient LAK cells, researches on activated NK cells for immunotherapy of glioma have been restricted. We pointed out a significant difference in the distribution of activated NK cells between low- and high-grade gliomas (p < 0.001) (Fig. 2c, d), and the lower the level of activated NK cells in the higher grade of gliomas (p < 0.001) (Fig. 3d). In studies of the association with glioma molecular subtype, the level of activated NK cells was the lowest in the mesenchymal subtype, which has the worst prognosis, while the other three subtypes harbor relatively higher level of it (p < 0.001) (Fig. 3g). In addition, as previously stated, the correlation coefficient of activated NK cells is also negative. These results indicate that activated NK cells may induce favorable clinical outcome of glioma, in another word, it may also be a vital suppressor for MT in LGG.

TAMs are macrophages infiltrating in tumor tissues which are the main composition in tumor microenvironment (TME). They differentiate through alternative pathways, among which the most common one is Notch pathway [26, 27]. What’s more, they facilitate tumor progression [28]. Once activated, monocytes continue to differentiate, first differentiated into M0 macrophages and then M1 and M2 arisen from M0. Others have shown that increased level of M0 is associated with poor clinical outcomes of lung adenocarcinoma [29]. So far, no clear experiments have been conducted to demonstrate the relationship between TAMs and glioma prognosis. Some people believe that TAMs in gliomas may be affected by tumor tissues and show immunosuppressive effects [19]. According to our work, the contents of M0 (p < 0.001) in GBM is higher than that of LGG. Wilcoxon test result also exhibits a gradual increase in the level of M0 from LGG to GBM. Moreover, coefficient of our IRS model also indicate that M0 comes under the influence of tumor development and promote malignant progression.

To again insight into the immune-related biological processes during glioma progression, we performed GO biological process (GOBP), GO cellular component (GOCC), GO molecular function (GOMF) and KEGG pathway analysis. Not only do the top results are immune-related, it is particularly worth mentioning that the first of each analysis are all immunologically relevant. This proves the validity of our consequences to some extent, on the other hand, it also finds some hub pathways in the MT of glioma, which indicates a path for future researches.

In summary, our study expounded the distinct composition of tumor-infiltrating immune cells in different grades and molecular subtypes of glioma. The complex intersection between TIICs and MT was quantified by our IRS model. Finally, we pointed out some relevant pathways related to progression and MT of glioma. These findings deepen the understanding of immune responses in CNS tumors and may enable to develop more effective immunotherapeutic strategies.