Integrated analysis of 34 microarray datasets reveals CBX3 as a diagnostic and prognostic biomarker in glioblastoma

Gliomas are the most common primary tumors in the central nervous system. According to the WHO criteria published in 2007 [1] and 2016 [2], gliomas are graded from I to IV, mainly including grade I–IV astrocytomas and grade II–IV oligodendrogliomas. Grade IV astrocytomas are known as glioblastoma (GBM), which is the most malignant and lethal glioma. GBM is characterized by high proliferation, infiltrative growth behavior, intratumoral heterogeneity and tumor recurrence. Despite improvements in GBM therapy that involve surgical resection, radiation and chemotherapy, a cure for GBM appears elusive. Additionally, the median survival is only 12–15 months for patients with glioblastomas [3]. The emergence of genomic and proteomic profiling has provided more insight into the oncogenesis, characterization, and therapy of gliomas.

The integration of molecular biomarkers with histological assessment has yielded new insights into gliomas [4‐6]. Some molecular biological markers are important for determining molecular subtypes, individualized treatment, and predicting prognosis, such as MGMT [7], EGFR [8], IDH [9], 1p19q [10], ATRX [11], MGMT promoter methylation levels, 1p/19q-codeleted and IDH1 mutations can predict the prognosis of GBM, oligodendroglioma (OD) and low grade glioma [12]. IDH1 or IDH2 mutations can exist in glioblastomas, especially evolved from lower-grade gliomas, and patients with such tumors had a better outcome than those with wild-type IDH genes [13, 14]. G-CIMP-positive status appears in most WHO grade II and III gliomas and secondary glioblastomas and is correlated with improved patient survival [15]. Other studies used EGFR, NF1, and PDGFRA/IDH1 to classify GBM into pro-neural, neural, classical, and mesenchymal subtypes [16, 17]. Although emerging evidence supports mRNAs as potential biomarkers of glioblastomas, gene expression studies analyzed in isolation usually have inconsistent or discrepant results. Various factors, such as limited sample sizes, different profiling platforms, and diverse methods for data collection and analysis, lead to these discrepancies. Furthermore, approximately half of all patients do not harbor known “driver” genes and cannot be treated with targeted agents. The NCBI Gene Expression Omnibus (GEO) contains numerous human microarray datasets from various types of tissue biopsies, which can be used to discover disease-associated biomarkers. These datasets represent a large and incompletely exploited resource for discovering novel biomarkers. However, the existence of biological (cohort selection) and technical (treatment protocol and microarray technology) differences in individual studies hindered the broader application of these findings and ultimately limited their translation into clinical practice. Thus, new approaches for the identification of novel biomarkers of gliomas are needed.

To overcome these limitations, we need an integrated and unbiased method of analyzing results and obtaining mRNAs with greater statistical significance. Through integrated analysis approaches, such confounding factors can be controlled by increasing the statistical power, thus allowing the detection of consistent biomarkers across multiple studies; such methods have been applied in analyses in breast cancer [18], prostate cancer [19], diffuse low-grade glioma [20], and lung cancer [21].

Among the gene expression signatures identified in our study, chromobox homolog 3 (CBX3/heterochromatin protein 1γ [HP1γ]), a member of the heterochromatin protein 1 (HP1) family, has the ability to regulate the structure of both heterochromatin and euchromatin, suggesting that it may participate in both transcriptional repression and activation [22]. CBX3 is associated with the epigenetic regulation of cell differentiation and cancer development [23]. Recently, CBX3 was revealed to be associated with lung cancer [24], osteosarcoma [25], gastric cardia adenocarcinoma [26] and colorectal cancer [27]. However, the precise role of CBX3 in glioma remains unclear, although Holmberg et al. [28] reported that HP1γ is associated with NPM1, which functions in the spatial organization of nucleolus-associated heterochromatin in glioma.

We performed integrated analyses of 34 gene expression datasets consisting of 1893 glioma samples, which enabled the discovery and validation of eight differentially expressed genes (DEGs) consistently and specifically expressed in GBM. We further validated aberrant expression patterns of eight DEGs in datasets from The Cancer Genome Atlas (TCGA) and the Chinese Glioma Genome Atlas (CGGA) and revealed that six DEGs are prognostic biomarkers of glioma. CBX3 was distinguished as a novel and clinically noteworthy mRNA in GBM. Further experiments showed that CBX3 was overexpressed in glioblastoma tissues and is a potential diagnostic and prognostic biomarker. Silencing CBX3 can arrest the cell cycle in the G2/M phase in U373 cells, thereby weakening the cell proliferation ability. We attempt to provide novel and critical biomarkers that might be beneficial for the precise diagnosis and prognostic prediction of GBM and have broader application for translation into clinical practice.

Many transcriptional studies in glioma have been performed; however, most used limited sample sizes, variable platforms and different sample types (cell or tissue), making it challenging to characterize stable and reliable molecular biomarkers of glioma. To our knowledge, our study is a very large integrated analysis to date of gene expression in glioblastoma. Eight genes were consistently expressed between GBM tissues and NG, A or OD tissues with high significance, a finding that may ultimately translate into clinical practice. To further assess our results, we investigated the role of these eight genes in the survival of patients with glioma using TCGA-GBMLGG and CGGA cohorts. Patients with high CBX3, BARD1, EGFR, or IFRD1 expression had significantly shorter survival and patients with high GUCY1A3 or MOBP expression had significantly longer survival than patients with lower expression of these genes.

Some of these eight genes, such as EGFR, STAT1, and BARD1, have been confirmed to be involved in cancer by numerous studies. Furthermore, some of these genes, such as MOBP and CTSS, have been confirmed to be related to glioma. CBX3, GUCY1A3, and IFRD1 have been reported in relation to some tumors but have been studied little in glioma.

MOBP is specifically overexpressed in oligodendrocytes [40]. Our study found that in the GBM vs. OD comparison, MOBP was overexpressed in OD, consistent with the literature. Thomas et al. [41] showed by an ELISA that CTSS was highly expressed in glioblastoma but was expressed at relatively low levels in grade I-III astrocytoma. In addition, high CTSS expression in glioblastoma shows a poor prognosis. This conclusion is completely consistent with our findings. GUCY1A3 is an upstream regulatory gene of VEGF and may be a molecular target for antiangiogenic therapy in glioma [42]. IFRD1 is expressed in various cells and may play a role in promoting tissue proliferation or regeneration [43].

However, research on IFRD1 in glioma has not been reported. Studies by Lewis and others showed that IFRD1 is highly expressed in colon cancer and is negatively correlated with the 5-year survival time. In our study, IFRD1 is specifically overexpressed in glioblastoma and suggests a poor prognosis in glioma.

In our study, CBX3 was upregulated in glioblastoma, and patients with high CBX3 expression had shorter survival. Considering that there is little research about CBX3 in gliomas, we selected CBX3 for further study. The mammalian HP1 family contains three isoforms, HP1a (CBX5), HP1b (CBX1), and HP1γ (CBX3). The CBX3-encoded protein HP1γ, a member of the heterochromatin family, is a highly conserved nonhistone chromatin protein containing two highly conserved domains. Current studies have confirmed that CBX3 is involved in transcriptional silencing, DNA repair, and RNA splicing. Moreover, the mechanisms of action of CBX3 in cancer remain obscure.

Han et al. [24] found that CBX3 was positively expressed in 90.3% of non-small cell lung cancer tissues, whereas only 2 of 7 normal lung tissues were positive for CBX3 expression. Saini et al. [25] identified that CBX3 can be used as a marker for tumor stem cells in osteosarcoma and that CBX3 was overexpressed in osteosarcoma and osteosarcoma metastases to the lung compared with its expression in primary osteoblasts. Liu et al. [27] found that CBX3 is overexpressed in colorectal cancer, while miR-30a is downregulated and inversely correlated with high CBX3 expression. Furthermore, CBX3 promotes colorectal cancer cell proliferation and tumorigenesis. p21 is a cyclin-dependent kinase inhibitor that can interrupt cell cycle progression, leading to cell cycle arrest [44, 45]. Knockdown of CBX3 increased p21 expression, resulting in slower proliferation of colorectal cancer cells. The miR-30a/CBX3/p21 axis is proposed to regulate the development of colorectal cancer and to be a prognostic and therapeutic target. Fan et al. [46] demonstrated that CBX3 can promote proliferation and cell cycle progression both in vivo and in vitro in colon cancer cells. CBX3 can promote the formation of colon cancer by inhibiting the expression of CDK6/p21, which are cell cycle (G1 phase to S phase) related genes. Several studies have also found that HP1 proteins interact with transcriptional regulators of key cell cycle genes, including cyclin E, E2F1, and p53 [47‐49]. In our study, CBX3 knockdown inhibited the proliferation of glioblastoma cells and led to cell cycle arrest at the G2/M phase to G0/G1 phase boundary, partly in accordance with the findings in the above studies.

However, there were also some limitations that should be strengthened in this study. First, since datasets in this study were from 15 different platforms, the batch effect is large, but we did quality control before DEG analysis to reduce the effect. Second, the functional analysis in wet experiment only explored CBX3, and further analysis in other gene were still needed.

In summary, this study is a global analysis identifying glioblastoma-specific mRNAs in such a large sample size through integrated analysis. Our analysis uses a “prevalidation” integrated analysis to identify signatures and wet lab experiments to validate the identified CBX3 gene, which may accelerate translational research and will provide insight into new strategies to seek tumor biomarkers for precision oncology.