This article is mentioned in:

Introduction

Gliomas are serious primary brain tumors arising from glial cells. According to the type of cells, they are histologically classified into astrocytomas, oligodendrogliomas and oligoastrocytomas. Gliomas are further categorized into grades I to IV, according to the World Health Organization (WHO) grading system (1). The most common form of gliomas are WHO grade IV tumors (glioblastoma and its variants), which account for 82% of cases of malignant glioma (2). Incidence rates of gliomas vary significantly with histological type, age at diagnosis, gender, ethnicity and nationality (3–7). High-grade gliomas typically recur after an average period of only 6.9 months, resulting in a median patient survival of just 12–15 months following diagnosis (8). Despite having an improved prognosis compared with increased grade glial tumors, 50–75% of patients with low-grading glioma also eventually succumb to the disease. Median survival times have been reported to range between 5 and 10 years, and estimates of 10-year survival rates range between 5 and 50% (9).

The last decade has witnessed a significant improvement in the understanding of the molecular mechanisms of glioma progress. For example, integrin inhibition leads to inactivation of the transforming growth factor (TGF)-β pathway, which controls features of malignancy, including immunosuppression, invasiveness and stemness in human glioblastoma (10). A previous study has reported that Crk-like protein efficiently regulates cell proliferation, migration and invasion induced by the TGF-β pathway in glioblastoma (11). Additionally, zinc finger E-box binding homeobox 1 expression is responsible for the invasion and chemoresistance of glioblastoma cells by regulating the downstream effectors roundabout, axon guidance receptor, homolog 1 (Drosophila), v-Myb avian myeloblastosis viral oncogene homolog and O-6-methyl-guanine-DNA methyl transferase (MGMT). Furthermore, previous studies have identified that epigenetic modifications occur in gliomas (12), and hypermethylated genes in gliomas include pithelial membrane protein 3, thrombospondin 1, excitatory amino-acid transporter 2 and MGMT (13–15). Based on the microarray dataset GSE31262, Sandberg et al have identified 423 upregulated and 414 downregulated genes in glioblastoma stem cells (GSCs) using the Rank Product algorithm, comparing adult human neural stem cells (ahNSCs), and inhibition of the Wnt-signaling pathway reduces proliferation and sphere forming capacity in GSCs; additionally, they also identified a 30-gene signature which is highly overexpressed in GSCs (16). Despite this, dysregulated gene networks and pathways associated with glioma progression remain undetected, and the molecular mechanisms of glioma remain elusive.

To further identify the potential key gene networks and pathways implicated in gliomas, based on the microarray data GSE31262 deposited by Sandberg et al (16), the differentially expressed genes (DEGs) were identified using the linear models for microarray data (limma) package, which is a popular choice for DEGs identification using microarray and high-throughput PCR data (17). Furthermore, co-expression interactions of DEGs were analyzed and a co-expression network was constructed. Functional and pathway enrichment analyses of DEGs in the co-expression network were also performed. Additionally, drugs closely associated with DEGs in the network were identified. The present study may provide more novel information for the investigation of the molecular mechanisms underlying gliomas, which contributes to a better understanding of the pathogenesis of gliomas.

Materials and methods

Obtaining and preprocessing of mRNA expression profile data

The mRNA expression profile dataset GSE31262 was downloaded from the GEO database (http://www.ncbi.nlm.nih.gov/geo/), which derived from 9 individual samples of GSCs and 5 individual samples of ahNSCs taken from the adult human brain (16). The annotation platform used here was GPL2986 ABI Human Genome Survey Microarray Version 2 (Applied Biosystems; Thermo Fisher Scientific, Inc., Waltham, MA, USA). The probe ID was first converted into a gene symbol. For probes being mapped to the same gene, the average expression value was calculated as the final expression value. Then, the data were translated into log2 logarithms, and the median normalization was performed using the robust multichip averaging method (18).

DEG screening

The limma package (19) in R language (http://www.bioconductor.org/packages/release/bioc/html/limma.html) was applied to identify DEGs between GSC samples and normal NSC samples. The false discovery rate (FDR) for each gene was obtained by adjusting the raw P-value with the Benjamini Hochberg method (20). FDR ≤0.01 and |log2 fold change (FC)| ≥1 were set as the thresholds.

Hierarchical clustering analysis of DEGs

To identify the sample-specificity of DEGs, hierarchical clustering (21) was conducted. Euclidean distance (22) was chosen as a measure of distance for DEGs between GSC samples and normal NSC samples. Pheatmap package (23) in R was used to visualize the result of hierarchical clustering.

Co-expression network of predicted target genes

An essential prerequisite for understanding cellular functions at the molecular-level is to correctly uncover all functional interactions among various proteins in the cells. The Search Tool for the Retrieval of Interacting Genes (STRING) database (http://string-db.org/) (24) was used to select co-expression interactions of DEGs, with a co-expression coefficient >0.6 as the cut-off. Based on the expression values of DEGs, the Pearson Correlation Coefficient (PCC) (25) between DEGs were calculated and only pairs with |PCC| >0.6 were retained. Then, the common pairs obtained by the two aforementioned methods were retained to construct the co-expression network, which was visualized using the Cytoscape software (http://www.cytoscape.org/) (26). In the network, a node represents a protein (gene), and lines represent the interactions of the proteins. The degree of each node represents the number of nodes that interact with this node.

Functional and pathway enrichment analyses

To reveal the functions of DEGs in the co-expression network, the online biological tool Database for Annotation, Visualization and Integrated Discovery (27) was used to perform Gene Ontology (GO) (28) functional enrichment analysis in biological process (BP). Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology-based Annotation System (KOBAS) was used to identify the pathways that were significantly enriched by genes. KOBAS provides the most comprehensive set of functionalities, including input by both IDs and sequences, identifying both frequent and statistically enriched pathways, offering the choices of four statistical tests and the option of multiple testing correction (29). P<0.05 was selected as the cut-off for the enrichment analysis.

Analysis of DEGs associated with glioblastoma prognosis

Preprocessed RNA-sequencing data and clinical data of glioblastoma were downloaded from the The Cancer Genome Atlas (TCGA) database (https://cancergenome.nih.gov/). Based on the clinical data, sample files that provided survival time and status of patients were chosen for subsequent analysis.

Cox proportional hazard model (30,31) was utilized to predict the association of DEGs and patient prognosis. In the present study, each DEG was defined as a covariant, and survival time of each chosen glioblastoma sample was defined as a dependent variable. Multivariate Cox proportional hazard regression analysis was performed for all DEGs. The obtained probability value for each covariant <0.005 was set as the cut-off, indicating that the DEG was associated with glioblastoma prognosis. The analysis was conducted using SPSS 19.0 software (IBM Corp., Armonk, NY, USA).

Identification of small-molecule drugs based on DEGs

Web-based Gene Set Analysis Toolkit (WebGestalt) incorporates information from different public resources and provides an easy way for biologists to make sense out of gene lists (32). The current version of WebGestalt includes disease and drug-associated genes. In the present study, drugs associated with genes in the co-expression network were obtained from WebGestalt. The raw P-value for the association between drugs and genes was calculated by the hypergeometric test, and adjusted using the multiple test adjustment. An adjusted P-value <0.05 was set as the cut-off.

Results

Identification of DEGs and hierarchical clustering analysis

Subsequent to preprocessing, standard DEG expression profile data was illustrated (Fig. 1). Based on the criteria of FDR <0.05 and |log2 FC| ≥1, a total of 431 DEGs were identified in the GSC samples, including 98 upregulated DEGs and 333 downregulated DEGs. Hierarchical clustering analysis illustrated that the GSCs samples and normal samples can be distinguished using the identified DEGs, according to their differential expression status (Fig. 2).

Figure 1. The boxplots for gene expression of each samples. (A) The boxplot showing gene expression of each sample prior to data normalization. (B) The boxplot showing gene expression of each sample subsequent to data normalization. The ordinate represents gene expression value, and the abscissa represents glioblastoma stem cell samples and normal controls. White columns represent normal samples, and gray columns represent glioblastoma stem cells samples. The X-axes represent the name of each sample, while the Y-axes represent the gene expression values of each sample and the horizontal lines in each column of the plot represents the mean gene expression value.

Figure 2. Hierarchical clustering analysis of differentially expressed genes. Each row represents a single gene; each column represents a sample. Red represents a high expression value; blue represents a low expression value.

Furthermore, comparing with DEGs that identified by Sandberg et al (16), 73 upregulated DEGs and 232 downregulated DEGs were common in the present study and the study by Sandberg et al (16) (Fig. 3).

Figure 3. Venn diagrams of differentially expressed genes. (A) Venn diagram of upregulated differentially expressed genes identified in the present study and the study by Sandberg et al (16). Up-1 represents the number of upregulated genes identified in the present study, and Up-2 represents the number of upregulated genes identified the study by Sandberg et al (16). (B) Venn diagram of downregulated differentially expressed genes identified in the present study and the study by Sandberg et al (16). Down-1 represents the number of downregulated genes identified in the present study, and Down-2 represents the number of downregulated genes identified in the study of Sandberg et al (16).

Co-expression network of the predicted target genes

To analyze the interactions of DEGs, STRING database and PCC were used to predict co-expressed interactions of DEGs. A total of 678 pairs of co-expression genes with co-expression score >0.6 were identified from the STRING database. In addition, 3,973 gene pairs with |PCC| >0.6 were extracted. Subsequent to comparison, the 100 common gene pairs were selected. The Cytoscape software was used to visualize the co-expression network (Fig. 4), which includes a total of 33 genes and 100 interactions. Notably, all genes in this network were upregulated, and the genes PDZ binding kinase (PBK), topoisomerase (DNA) II α (TOP2A), cyclin dependent kinase (CDK) 1, cell division cycle (CDC) 6 and NIMA related kinase 2 (NEK2) had a higher degree that was >10. PBK, TOP2A and CDK1 were co-expressed with each other.

Figure 4. Co-expression network of differentially expressed genes. Triangles represent upregulated differentially expressed genes.

Functional and pathway enrichment analyses of genes in the co-expression network

To further reveal the functions of DEGs in the co-expression network, the GO functional and KEGG pathway enrichment analyses were conducted. The majority of the GO BP terms were associated with the cell cycle, including cell cycle procession, nuclear division and chromosome segregation (Table I). Additionally, two significantly enriched pathways (cell cycle and p53 signaling pathway) were obtained (Table II). The upregulated genes involved in the cell cycle pathway included cyclin B1 (CCNB1), CDK1, CDC6, CDC45, DBF4, proliferating cell nuclear antigen, cyclin A2 and CDK2. Genes enriched in the p53 signaling pathway were CCNB1, CDK1, ribonucleotide reductase regulatory subunit M2 and CDK2.

Table I. Biological processes of differentially expressed genes in the co-expression network.

Table I.

Biological processes of differentially expressed genes in the co-expression network.

Term	Process	Count	P-value	FDR
GO:0022403	Cell cycle phase	21	2.02×10−23	2.79×10−20
GO:0007049	Cell cycle	24	1.03×10−22	1.42×10−19
GO:0000278	Mitotic cell cycle	20	1.31×10−22	1.80×10−19
GO:0022402	Cell cycle progression	22	2.23×10−22	3.08×10−19
GO:0000279	M phase	19	8.50×10−22	1.17×10−18
GO:0007067	Mitosis	17	3.30×10−21	4.56×10−18
GO:0000280	Nuclear division	17	3.30×10−21	4.56×10−18
GO:0000087	M phase of mitotic cell cycle	17	4.43×10−21	6.12×10−18
GO:0048285	Organelle fission	17	6.35×10−21	8.77×10−18
GO:0007059	Chromosome segregation	10	1.13×10−13	1.57×10−10
GO:0051301	Cell division	13	9.02×10−13	1.25×10−9
GO:0051726	Regulation of cell cycle	13	3.51×10−12	4.85×10−9
GO:0006260	DNA replication	10	2.72×10−10	3.75×10−7
GO:0051276	Chromosome organization	11	7.20×10−8	9.95×10−5
GO:0006259	DNA metabolism	10	1.29×10−6	1.78×10−3

[i] FDR, false discovery rate.

Table II. Pathway enrichment analysis of the differentially expressed genes enriched in the co-expression network.

Table II.

Pathway enrichment analysis of the differentially expressed genes enriched in the co-expression network.

ID	Pathway	P-value	FDR	Genes
hsa04110	Cell cycle	6.96×10−9	1.39×10−7	CCNB1, CDK1, CDC6, CDC45, DBF4, PCNA, CCNA2, CDK2
hsa04115	p53 signaling pathway	5.94×10−4	5.93×10−3	CCNB1, CDK1, RRM2, CDK2

[i] FDR, false discovery rate.

Screening of DEGs associated with glioblastoma prognosis

To further analyze whether the identified DEGs were associated with the prognosis of patients with glioblastoma, the survival data from the TCGA database were used in the present analysis. Among the sample files obtained from TCGA, a total of 107 files provided survival time and status of patients. Based on the Cox proportional hazard regression analysis, 69 DEGs identified in the present study were significantly associated with glioblastoma prognosis, 4 of which were present in the co-expression network, including CDK2, kinesin family member 15, kinesin family member 18A and TOP2A.

Identification of small-molecule drugs based on DEGs

To investigate potential drugs that may be used to treat human glioblastoma, WebGestalt was used in the present study to identify the drugs associated with the DEGs in the co-expression network. A total of 9 drugs were obtained, including paclitaxel (for genes including TOP2A, BIRC5 and CDK1) and etoposide (for genes including TOP2A and CDC6) (Table III).

Table III. The significant drugs associated with differentially expressed genes in co-expressed genes.

Table III.

The significant drugs associated with differentially expressed genes in co-expressed genes.

Drug	ID	Parameters	Genes
Paclitaxel	DB_ID:PA450761	O=5; rawP=9.99×10−9; adjP=1.90×10−7	TOP2A, BIRC5, CDK1, TACC3, CENPF
Etoposide	DB_ID:PA449552	O=5; rawP=4.20×10−8; adjP=3.99×10−7	TOP2A, CDC6, BIRC5, CDK1, DBF4
Hydroxyurea	DB_ID:PA449942	O=4; rawP=1.86×10−7; adjP=8.83×10−7	RRM2, PCNA, CDK2, RAD51
Mechlorethamine	DB_ID:PA450336	O=3; rawP=1.85×10−7; adjP=8.83×10−7	CCNB1, CDK2, CDK1
Podofilox	DB_ID:PA450993	O=4; rawP=7.57×10−7; adjP=2.88×10−6	TOP2A, CDC6, BIRC5, DBF4
Progesterone	DB_ID:PA451123	O=4; rawP=3.61×10−6; adjP=9.80×10−6	TOP2A, CCNA2, CCNB1, RAD51
Cisplatin	DB_ID:PA449014	O=4; rawP=3.50×10−6; adjP=9.80×10−6	TOP2A, RRM2, BIRC5, RAD51
Docetaxel	DB_ID:PA449383	O=3; rawP=4.93×10−6; adjP=1.17×10−5	TOP2A, RRM2, BIRC5
Adenosine	DB_ID:PA448049	O=5; rawP=2.98×10−5; adjP=5.66×10−5	TOP2A, PCNA, CDK2, RAD51, NCAPG

[i] O, number of genes in the gene set and in the category; rawP, P-value from hypergeometric test; adjP, P-value adjusted by the multiple test adjustment.

Discussion

Malignant gliomas are the most frequently occurring primary brain tumors, and they have a poor survival rate (8). The present study aimed to investigate the potential crucial genes correlated with glioblastoma. Based on the microarray data, a total of 98 upregulated DEGs and 333 downregulated DEGs were identified from the GSCs, of which, 33 upregulated genes had co-expression interactions. In the co-expression network, genes including PBK, TOP2A, CDK1, CDC6 and NEK2 had a higher degree. A set of genes including CCNB1, CDK1, CDC6 and CDK2 were significantly enriched in the cell cycle.

The PBK gene encodes PBK, and it has been previously suggested that mitotic phosphorylation is required for its catalytic activity (33,34). The present results illustrated that PBK expression is upregulated in gliomas. Multiple previous studies have shown the present results (35–37). A previous study has demonstrated that gene knockdown of PBK leads to decreased viability and sphere formation of glioma initiating cells (GICs), and the PBK inhibitor HITOPK-032 abolishes growth and elicits a large increase in apoptosis of GICs, indicating an important role of PBK in the cell growth of glioma (38). As a result, the present study speculated that PBK may play a pivotal role in glioblastoma via the regulation of the cell cycle. The TOP2A gene encodes TOP2A which controls and alters the topologic states of DNA during transcription (39). The present results showed an upregulated expression of TOP2A protein in glioma, which has also been identified in numerous cancers (40–42). In addition to its roles in DNA synthesis and transcription, and chromosomal segregation during mitosis, Tsavaris et al (43) has reported that TOP2A is also a sensitive and specific marker of actively proliferating cells. In the present study, TOP2A was found to be significantly involved in the prognosis of patients with glioblastoma, which was consistent with numerous previous studies (44–46). NEK2 encodes a serine/threonine-protein kinase, which is involved in mitotic regulation (47). Laakso and Hautaniemi (48) has previously reported that NEK2 is upregulated in glioblastoma, which is consistent with the present findings. The overexpression of NEK2 has also been found in human breast and liver cancer, and colorectal carcinoma (49). Cappello et al (50) identified that NEK2 knockdown induced aneuploidy and cell cycle arrest that further led to cell death in various human breast cancer cell lines. Therefore, inhibiting the expression of the PBK, TOP2A and NEK2 genes may be a novel approach for glioma therapy.

According to the pathway enrichment analysis, the majority of significantly altered pathways were involved in the cell cycle, which is consistent with the BP analysis, and genes involved in this pathway included CDK1, CDC6, DBF4, CCNB1 and CDK2. Cancer cells differ from normal cells by numerous important characteristics, including loss of differentiation, increased invasiveness and decreased drug sensitivity. The conversion of normal cells into cancer cells is facilitated by the loss of fidelity in cell cycle progression, which under normal conditions is achieved by the coordinated activity of CDKs, checkpoint controls and repair pathways in the cell cycle (51–53). Furthermore, numerous studies have shown that this fidelity can be abrogated by specific genetic changes (54–56). Both CDK1 and CDK2 encode CDKs, which are highly conserved proteins functioning as a serine/threonine kinase (57,58). Overexpression of CDK1 and CDK2 promotes oncogenesis and progression of human gliomas, whereas downregulation of CDK1 and CDK2 expression inhibits the proliferation activities of human malignant gliomas (59,60). Furthermore, CDK2 was identified as a prognostic gene in the present study, which was consistent with the study by Zhang et al (61). Additionally, CDK2 is involved in the expression of other prognostic markers, including p27 and c-Met (62,63). CDC6 encodes CDC6, which is a protein essential for the initiation of DNA replication (64). It has been demonstrated that CDC6 expression is a marker of proliferative activity in brain tumors, including gliomas (65). CCNB1 encodes CCNB1, which is a regulatory protein of cdk1 in mitosis (66). An abnormally high expression level of CCNB1 has been reported to be correlated with high grades and advanced stages of gliomas (67,68). Collectively, genes such as CDK1, CDK2, CDC6 and CCNB1 may exert critical roles in the progress of glioblastoma through mediating cell cycle.

In the present study, CDK1, CDK2 and CCNB1 were also markedly enriched in the p53 signaling pathway. P53 signaling pathway is the most commonly mutated pathway in tumorigenesis (69). As a well-known tumor suppressor, p53 responds to DNA damage and various genotoxic and cytotoxic stresses by inducing cell cycle arrest and apoptosis (70). Overall, the p53 pathway is disrupted in ~80% of high-grade gliomas (WHO Grades III and IV) (71). Therefore, in the progress of glioblastoma, CDK1, CDK2 and CCNB1 may also regulate the p53 signaling pathway.

Additionally, the present study used the limma package to identify DEGs, comparing the method (Rank Product algorithm) used by Sandberg et al (16). Among the identified DEGs in the present study, 73 upregulated genes and 232 downregulated DEGs were common with those screened by Sandberg et al (16), indicating the potentially significant roles of these common DEGs in glioblastoma. However, another 25 upregulated and 101 downregulated DEGs, which were not found by Sandberg et al (16), were also screened. The results suggest that it is necessary to reanalyze the public microarray data using different methods. The roles of these DEGs identified in the present study in glioblastoma require additional investigation. Additionally, the expression levels and functions of the genes associated with the cell cycle and the p53 signaling pathway in glioblastoma require validation by experiments, which will be conducted and reported later.

In conclusion, a total of 98 upregulated DEGs and 333 downregulated DEGs were identified from the GSCs compared with the ahNSCs. The upregulated hub genes in the co-expression network (including PBK, TOP2A, CDK1, CDC6 and NEK2) and the DEGs (including CCNB1, CDK1, CDC6, and CDK2) that correlated with cell cycle and the p53 signaling pathway may have essential roles in the progress of glioblastoma. These genes may be potentially therapeutic targets of gliomas, and the findings may be helpful to identify the molecular mechanisms of the pathogenesis of glioblastoma.

References