Immune genes are associated with human glioblastoma pathology and patient survival

Glioblastoma multiforme (GBM) is the most common and aggressive primary brain tumor in adults. Despite recent advances in multimodal therapy, prognosis remains limited [1]. Conventional treatment, generally maximal safe surgical resection followed by combination radiation and chemotherapy with temozolomide, fails to prevent tumor recurrence.

Recently, molecular subtypes of brain tumors have been characterized by microarray gene expression profiles [2–6]. These subgroups have been associated with significant differences in tumor aggressiveness, progression, and/or prognosis [7]. Gene expression analysis has been reported as being more accurate than conventional histology [8, 9]. Due to this greater accuracy, expression-based classifications offer an opportunity to improve molecular classification of gliomas [6, 7] and clinical diagnosis of glioblastomas [2]. Such advances could be helpful in designing future therapeutic trials [4, 10].

Many arguments have supported a link between the immune system and glioma pathogenesis. In several epidemiologic studies, glioma incidence is inversely associated with allergy history [11–13]. T-lymphocyte infiltration has been reported in certain glioma patients and an elevated number of intratumoral effector T cells has been recently correlated with a better survival in GBM patients [14].

Interestingly, several transcriptomic studies using microarray technologies have also reported an immune signature in gene expression profiling of glioma [8, 10, 15, 16] and GBM [17–20]. A signature associated with myeloid/macrophagic cells has been reported in most of these studies [10, 15, 16, 18, 20], a finding consistent with the known macrophage/microglia infiltration in GBM [21–23]. More recently, transcriptomic studies in glioma have revealed different signatures involving immune genes associated with overall survival (OS) [8, 10, 15, 19]. Gravendeel et al. reported an immune response signature associated with poor survival in glioma (Cluster 23 – the M function category) [8]. Murat et al. reported better outcome in patients with gene clusters characterizing features of innate immune response and macrophages (G24 cluster – 134 probes, among them probes for CD11b and CD163 genes) [19]. In contrast, Irliev et al. found an immune module (M7 module) associated with short survival that includes 449 genes, among them T-cell markers (CD4, CD8) and myeloid markers (MHC class II, TLR1 and TLR2) [15]. An NK cell signature (G12 gene cluster including Fc gamma receptors and DAP-12) has previously been reported in one study with higher level expression in primary GBM with shorter survival compared to low grade astrocytomas and secondary GBM [10].

In order to clarify the possible role of immune cells in GBM pathology and OS, we have performed a co-expression network analysis focusing on 791 genes linked to the immune system. Using a meta-analysis approach and independent validation cohorts, we identified an immune signature of GBM linked to innate immunity involving myeloid and NK cells as well as a 6-immune genes risk-model stratifying patients into two groups with significantly different OS.

In this study, we were able to link IA genes expression pattern with GBM biology and patient survival. Indeed, our co-expression network analysis highlighted clusters of IA genes and revealed related immune signatures marking innate immunity, NK and myeloid cells and cytokines/MHC class I molecules profiles. Furthermore, 108 IA genes were associated with OS. Among these, 6 IA genes were included in a weighted multigene risk model that can predict outcome in GBM patients.

Several studies have previously reported an immune signature in GBM [8, 10, 15–17, 19, 20, 30]. A signature associated with myeloid/macrophagic cells was reported in most of these [10, 15, 16, 18, 20]. We also found such a signature linked to one co-expression module for which annotation enrichment found monocytes, leukocyte activation and macrophage-mediated immunity. The well known macrophage/microglia infiltration in GBM can account for up to one-third of cells in some GBM specimens [21–23]. Unlike Ivliev et al. [15], we were unable to identify a T-cell signature in our analysis. Nevertheless, the association of two gene modules with GBM patient survival suggests that innate immunity including NK cell functions and cytokines/CMH class I profiles might affect outcome in GBM patients. A NK cell signature has previously been reported in one study in primary GBM [10]. NK cell infiltration was described earlier in glioma [31] but was not confirmed by others [32]. It is noteworthy that in murine glioma models, various vaccines strategies using CCL2 [33], CpG [34], IL12-expressing stroma cells [35] or IL23-expressing dendritic cells [36], induced an increased recruitment of NK cells at the tumor site, associated with better overall survival.

Most of chemokines present in the cytokines/MHC class I module are involved in recruiting T cells, monocytes/macrophages and neutrophils: e.g. CX3CR1/CX3CL1, CXCL9 and CXCR2 genes. In addition, most of the cytokines found such as MIF, IL5, IL12A and IL16 genes are known to regulate macrophages/monocytes, eosinophils, NK and T cells. Lohr has also reported that intratumoral infiltration of effector T cells is associated with a better survival in GBM [14]. In total, one could speculate that these two modules associated with overall survival reflect the recruitment and activation of immune cells such as NK cell, T cell, macrophages/monocytes, or neutrophils that would interfere with GBM patients’ survival. Interestingly, several clinical trials using dendritic cells have reported that the presence of T cells and neutrophils at the tumor site is associated with longer survival of the vaccinated patients [37]. Recently, Ducray et al. reported that infiltration of both CD3+ T cells and CD68+ macrophages was observed more frequently in GBM responders than in non-responders to radiotherapy [17]. However, in the present study, we did not find any association between key regulators of the T cell biology such as GATA3, TBX21 (TBET), and RORC (ROR-gamma-t) with patients’ survival (data not shown). The small amount of these infiltrating cells is usually reported in the GBM specimens and might have impaired the identification of such genes by a transcriptomic approach.

In addition to the co-expression network analysis, we have found 108 IA genes directly associated with OS in GBM patient using three different statistical methods. These genes are known to be involved in the biology of B cells (i.e. immunoglobulins, BLNK, CD19, CD20 and CD22 genes), T cells (i.e. CD1E, PTCRA, CD247), NK cells (i.e. KIR2DL1, KIR2DL4 and KIR3DL3 genes), and myeloid cells including monocytes/macrophages (i.e. ADAMDEC1, CD89/FCAR, CD64/FCGR1B and FCGR1C genes) and neutrophils (i.e. CD89, and NCF1B genes). Surprisingly, other important genes expressed by glioma-infiltrating microglia/macrophages, such as CD163 and AIF1 (IBA1), were not significantly associated with patients’ survival (data not shown). Komohara et al. have recently reported that the presence of CD163+ CD204+ M2-type macrophagic cells correlates with glioma grading and survival using an immunohistochemistry approach [38]. This discrepancy between our results and the Komohara et al. study could be explained by the fact that we used different technical approaches to detect these markers: at the mRNA level in our genomic study and at the protein level in [38]. Others genes of chemokines and cytokines have been also found such as CCL15, CCL17 IL1B and IL5 genes. Finally, some genes are known to be involved in the modulation/suppression of the immune response such as APRIL, ARG1, CD70, B7-H4, ICOSLG, NOS2A, TGFB1 and TWEAK genes.

Finally, we have developed a 6-IA-gene risk predictor of OS in GBM patients. The genes have been selected for an optimal survival model built on IA genes associated with survival as described in de Tayrac et al. [2]. This 6-IA gene risk is able to discriminate patients treated by chemo-radiation therapy into two distinct groups with significantly different survivals. These genes ACVR2A, ARG1, CD22, FGF2, MNX1 and RPS19 were present in all but one of the co-expression modules. The ‘regulation of immune response’ module, which contains no gene retained in the 6-IA-gene risk predictor, is the only one that does not include survival-associated genes. ACVR2A, CD22 and MNX1 genes were found to be associated with GBM patient survival in the three different statistical methods. Intriguingly, these 6 IA genes are not specific markers for known immune cell subpopulations. They are involved in the activation or the inhibition of the immune system. As a result, they impact positively or negatively on the risk predictor. For example, the expression of ARG1, a gene involved in immunosuppression, contributes positively to the 6-IA-gene risk index and therefore decreases the patient’s probability of survival. Although these genes are known in other cancers, they have not been described in GBM. ACVR2A is a receptor for activin-A and controls cell proliferation [39], for example proliferation of prostate cancer cells [40]. Mutations of ACVR2A are commonly found in unstable colonic cancers [41], and interestingly, infiltration of CD3 T cells is associated with mutated ACVR2A genes [42]. ARG1 for arginase-1 is a cytosolic enzyme that hydrolyses arginine to urea and ornithine [43]. ARG1 has recently been involved in immunosuppressive mechanisms by reducing T-cell activation [44]. CD22 cannot be considered only to be a B cell receptor that mediates cell adhesion and signaling [45, 46] since Mott et al. report that neurons can secrete this molecule [47]. Neuronal secretion of CD22 inhibits microglia activation via interaction with CD45 [47]. FGF2 for fibroblast growth factor-2 stimulates GBM growth [48]. Nevertheless, the high molecular weight FGF2 isoform inhibits glioma proliferation [49] and explains the radiation therapy resistance pathway [50]. Interestingly, plasma levels of FGF are higher in GBM patients compared to control [51]. MNX1 gene is involved in a congenital malformation, the Currarino syndrome (congenital malformation) [52] and also previously reported in CD34+ cells, B cells and B lymphoid tissues [53]. MNX1 function in immune cells and GBM biology has not been demonstrated yet but it has recently been described as a transcriptional factor implicated in the development of both solid and hematological cancers [54]. RPS19 is a subunit of 40S ribosome involved in pre-rRNA processing but also has extra-ribosomal functions. Indeed, RPS19 can act as a chemokine that regulates macrophage migration inhibitory factor (MIF) negatively [55]. Moreover, RPS19 can interact with FGF2 to drive differentiation or proliferation pathways of various cell types [56]. Only one statistical method, the quartile method, found this gene significantly (Figure 3), but the co-expression module found it to be significantly associated with OS (Figure 2).

To validate the strength of our 6-IA-gene risk predictor, expression of these genes was tested in a local cohort using RT Q-PCR. This technique has at least two advantages, it is used routinely in most laboratories and is relatively inexpensive compared with genomic microarray technologies. The test cohort was small (57 GBM specimens) but homogeneous in terms of treatment: combined surgery and chemo-radiation therapy [1]. In addition, the MGMT methylation status, which is the best predictor of response to the current combination treatment, was determined for all GBM specimens. Applied to this small cohort, 6-IA-gene risk predictor was even able to discriminate significantly between patients with high and low risk in the good prognosis group, defined by methylation of the MGMT promoter.

Recent advances in glioma classification have been achieved using genomic analysis. It is now accepted that GBM can be categorized in four subtypes defined as proneural, neural, mesenchymal, and classical groups [6, 7, 24]. The clinical outcome of the patients is different according to the GBM subtype. For instance, patients with proneural subtype live longer and the standard treatment does not increase their overall survival [6, 7]. In contrast, overall survival of patients with classical or mesenchymal subtype is significantly increased with the standard treatment. Interestingly, we have shown that our 6-IA-gene risk predictor was powerful in GBM proneural subtype but not in others subtypes. GBM proneural is an atypical GBM subtype which is associated with younger age, PDGFRA gene amplification, IDH1 mutations, TP53 mutations [7]. Due to the fact that these patients with proneural GBM have longer survival, one could speculate that the anti-tumor immune response could have more time to occur and slow down the tumor progression in some of these patients with a particular immune profile, revealed by our 6-AI-gene risk predictor.

In conclusion, we have demonstrated that GBM are characterized by an immune signature which could reflect the infiltration and activation of immune cells or the immunosuppression mechanisms developed by the tumor itself. Several IA genes were found to be associated with clinical outcome of GBM patients, allowing us to describe a 6-IA-gene risk predictor. This risk model can discriminate between patients with different outcomes, even within the good prognosis group based on MGMT status and within the proneural GBM subtype group. Further studies are needed to understand how these IA genes are involved in the control of GBM progression. Overall, this study highlights the important role of the immune system in the battle against the tumor and suggests new strategies for further development of immunotherapy for GBM patients.

This work was supported by Gefluc Rennes Bretagne (Groupement des Entreprises Françaises dans la Lutte contre le Cancer) et la Région Bretagne (CPER «  Biothérapies » ).