Glioblastoma (GBM; grade IV astrocytoma) is a very aggressive form of brain cancer with a poor survival and few qualified predictive markers. This study integrates experimentally validated genes that showed specific upregulation in GBM along with their protein-protein interaction information. A system level analysis was used to construct GBM-specific network. Computation of topological parameters of networks showed scale-free pattern and hierarchical organization. From the large network involving 1,447 proteins, we synthesized subnetworks and annotated them with highly enriched biological processes. A careful dissection of the functional modules, important nodes, and their connections identified two novel intermediary molecules CSK21 and protein phosphatase 1 α (PP1A) connecting the two subnetworks CDC2-PTEN-TOP2A-CAV1-P53 and CDC2-CAV1-RB-P53-PTEN, respectively. Real-time quantitative reverse transcription-PCR analysis revealed CSK21 to be moderately upregulated and PP1A to be overexpressed by 20-fold in GBM tumor samples. Immunohistochemical staining revealed nuclear expression of PP1A only in GBM samples. Thus, CSK21 and PP1A, whose functions are intimately associated with cell cycle regulation, might play key role in gliomagenesis. Cancer Res; 70(16); 6437–47. ©2010 AACR.

Glioma occurs at an incidence of almost 6 to 12 per 100,000 people globally and approximately 70% to 80% of all primary malignant brain/central nervous system tumors are of glial origin of which astrocytoma accounts for ∼60%. The most recently adapted WHO classification describes four grades of astrocytomas based on biological potential and ascending scale of aggression: grade I (pilocytic astrocytoma), grade II, [low-grade diffuse astrocytomas (DA)], grade III [anaplastic astrocytoma (AA)], and grade IV glioblastoma (GBM; ref. 1). The prognosis of patients with GBM is dismal, and the median survival of patients is only 10 to 12 months. The development of GBM proceeds through two genetic pathways resulting in primary and secondary GBM. The most frequently presenting form of GBM is the primary GBM, which develops de novo without any history of a less malignant precursor. These patients are normally of older age (>40 y) and are characterized by a high rate of EGFR gene amplification, p16 INKA deletion, mutations in PTEN gene, and MDM2 gene amplification (2). On the other hand, the secondary GBMs develop after a preceding stage of lower grade of astrocytomas, characterized by TP53 and IDH1mutations, RB gene alterations, and amplification and overexpression of PDGFR gene. A recent study by Ruano and colleagues (3) showed the existence of a minority subgroup of primary tumors with concurrence of epidermal growth factor receptor (EGFR) and p53 aberrations resulting in worse clinical outcome owing to genomic instabilities and cell cycle deregulation, indicating that an overlapping mutation spectrum might exist in a subset of gliomas.

Until recently, the only distinctive pathologic classification came from histologic interpretations of the tumor sample. Recent technique of microarray-based genome wide expression profiling has provided exciting opportunity to identify genes that are differentially regulated in different types of cancers (46). Several studies have described the identification of both upregulated and downregulated genes in primary and secondary GBM (79). Gene signatures have been identified, which can serve as potential markers associated with tumor grade, progression, and also patient survival (10). More recently, molecular categorization of GBM based on predictive roles of genes such as EGFR, NF1, and PDGFRA/IDH1, each defining the classic, mesenchymal, and proneural subtypes has been proposed (11, 12). Chuang and colleagues (13), using systems biology approach, have concluded that protein network-based classification of metastasis in breast cancer is more robust and reliable than gene signatures alone. This approach is based on the coherent expression patterns, conservation of subnetworks across species that fall into distinct pathways. The advantages of this method are (a) better understanding of molecular mechanisms, (b) identification of highly connected expression responsive genes otherwise not detected by differential gene expression, and (c) enhanced reproducibility between different patient cohorts than individual markers. This network approach is greatly facilitated by the available literature on protein-protein interaction data validated by either yeast two-hybrid system or mass spectrometry. Horvath and colleagues (14), recently using this network approach, identified abnormal spindle-like microcephaly (ASPM associated, a centrosome-binding protein) as an important key molecule involved in the progression of GBM.

In this communication, we have used the combined approach of differential gene expression and protein-protein interaction data to generate a network of GBM-specific upregulated genes. In this exercise, we have considered those genes that were experimentally verified to be upregulated, which served as seed proteins. Intricate connections within this network were then dissected to identify novel connecting links that united several hub proteins and could serve as molecular markers for prognostic value. Here, we describe the identification of two proteins namely protein phosphatase 1 α (PP1A) and CSK21 as important connecting molecules of two subnetworks involving key players involved in cell cycle regulation. The expression pattern of these two markers were further tested by quantitative reverse transcription-PCR (RT-PCR) and immunohistochemistry (IHC) in GBM patient tumor samples.

Network data set

We curated gene expression literature published until 2007 for GBM using various key words such as (a) cancer, microarray, GBM, and human; (b) primary and secondary GBM, (c) GBM prognostic markers, (d) glioma meta-analysis; and (e) glioma and microarray. Normalized microarray data sets from patient's GBM tumor sample compared with those of nonneoplastic brain samples were taken for further analysis. Many of these gene expression studies embody descriptions of both upregulated and downregulated genes. Albeit, we observed that most of these studies have attempted to validate further only the overexpressed but not the downregulated genes using biochemical methods (15, 16). Hence, we took only upregulated genes for our analysis. This exercise yielded a list of 124 genes that were statistically significant (P ≤ 0.05) and showed >2-fold change in gene expression. We filtered this list based on experimental support in the literature by at least one of the biochemical methods, which gave us 62 genes that served as input for our network construction (Supplementary Table S1).

Network modeling and randomization

We collected experimentally validated protein-protein interaction data from the APID database for each of the seed proteins using the Cytoscape software v2.6.1 (17, 18). The input lists were seed proteins, and the reported interacting proteins were neighbors connected to each other through links/edges. We used Erdos-Renyi model denoted by G (n,p) to obtain randomized network as control, keeping the node degree constant as that of specific networks. We queried, for a given value of node degree n similar to that of specific GBM networks, what is the probability that they are connected. The edges were shuffled 1 × 104 times before analyzing the topological features of the networks thus generated.

Analysis of global properties and statistical analysis of networks

The topological characteristics of network were analyzed by the Network Analyzer and CentiScape software (19, 20), with respect to node degree distribution, power law function, clustering coefficient, average path length, topological coefficient, and betweeness. To measure similarity and diversity in subnetworks, we calculated the Jaccard index using the number of shared interactors between two individual hubs (21).

JaccardIndex=|InteractorsofproteinAInteractorsofproteinA||InteractorsofproteinAInteractorsofproteinA|.

Significance of 3, 4, and 5 component modules was deduced based on the number of overlapping interactors shared between two or more proteins. Z-score (Z) was calculated while keeping the threshold value of more than 2 for the module (21).

Z=xx̄σ.

Pathway enrichment analysis for subnetwork proteins was done using the Genotypic Technologies Biointerpretor tool.

Normal brain and GBM tumor samples for experimental validation

Samples of diffusely infiltrating astrocytoma of different grades (DA grade II, AA grade III, GBM grade IV) were obtained from patients who underwent surgery at the National Institute of Mental Health and Neurosciences and Sri Satya Sai Institute of Higher Medical Sciences, Bangalore, India. A portion of anterior temporal cortex resected from patients who underwent surgery for intractable epilepsy served as control sample. RNA extracted from 7 control samples, 8 DA, 26 AA, and 44 GBM samples were used to examine the expression of CSK21 and PP1A genes in triplicate by quantitative real-time PCR (9). The genes ATP5G1, AGPAT1, RPL35A, and GARS were used as internal controls (see Supplementary Table S2 for primer sequences; ref. 22). Δ-Δ Ct method was used for the relative quantification of gene expression. Statistical significance was tested by ANOVA using the GraphPad PRISM software. Log2 ratios were calculated for each group and plotted as scatter plots depicting the median value of each subgrade.

Immunohistochemistry

Paraffin sections (4 μm) from the tumor tissue and control samples were collected on silane-coated slides, and the protein expression of PP1A and CSK21 was assessed by IHC on 25 samples (4 DA, 5 AA, 13 GBM, and 3 control samples). The heat-induced antigen retrieval of the deparaffinized sections was performed as described (9). Sections were incubated overnight with primary antibody as follows: PP1A (Cell Signaling) at 1:25 dilution and CSK21 (C-19: sc-6477, Santa Cruz Biotechnology) at 1:50 dilution overnight at room temperature. This was followed by incubation with the streptavidin-linked biotinylated secondary antibody (universal LSAB, DAKO) for CSK21 and with supersensitive horseradish peroxidase detection system (QD440-XAK, Biogenex) for PP1A. 3, 3′-Diaminobenzidine (Sigma-Aldrich) was used as the chromogenic substrate. Tumors that showed markedly increased mRNA levels of PP1A and CSK21 served as positive controls. A negative control slide in which the primary antibody is omitted was included with each batch of staining. Both nuclear and cytoplasmic staining was noted for PP1A and CSK21. The nuclear and cytoplasmic immunohistochemical staining were scored semiquantitatively on a three-point scale of 0 to 2, in which 0 is no staining, 1+ is mild staining, and 2+ is strong staining within the tumor core. Only 2+ nuclear and cytoplasmic positivity were considered for analysis. The immunopositivity was assessed in >1,000 cells from each tumor specimen. The labeling index (LI) was expressed as percentage of cells that showed 2+ staining among the total number of cells counted.

Statistical analyses for IHC data

All continuous variables were tested for normal distribution, and they were found to be nonnormal. To determine the grade-specific expression pattern, a nonparametric test, Kruskal-Wallis One-Way ANOVA on Ranks, was performed on the data assessed by IHC, for testing equality of population medians among groups. Significance in the differences between the mean of all of the four groups was determined by one-way ANOVA followed by the post hoc Tukey's Honestly Significant Difference test for multiple comparisons to adjust the α level for minimizing the type I error rate (23, 24). The results were expressed in the form of mean LI ± SD. All analyses were performed using the SPSS ver. 15.0 (SPSS, Inc.).

GBM-specific network construction and its properties

Our extensive literature search revealed that 124 genes are upregulated in GBM among which 62 genes were found to be overexpressed (P ≤ 0.05; fold change, >2) in microarray expression data sets, which were also verified by at least one of the following techniques such as RT-PCR, qRT, IHC, and tissue microarray. These genes were taken further as seed proteins for constructing GBM-specific network (Supplementary Table S1). We extracted their protein-protein interaction information through the APID2NET software (17). Network construction methods is presented in Fig. 1, and the connections thus made between all the proteins were defined as edges or links. Such an exercise of network construction of the seed proteins resulted in the formation of a large network with 11,752 interactions between 1,447 proteins visualized using the Cytoscape v 2.6.1 software (Supplementary Fig. S1; ref. 18). As a control, we also created randomized network by shuffling the edges 1 × 104 times using the same number of protein-protein interactions as were there in GBM network.

Figure 1.

The flowchart describes the construction of the GBM network by integrating gene expression and protein-protein interaction data followed by calculation of topology parameters such as number of nodes, number of edges, average node degree, average clustering coefficient and shortest path of the network, and identification of subnetworks and their novel interactions.

Figure 1.

The flowchart describes the construction of the GBM network by integrating gene expression and protein-protein interaction data followed by calculation of topology parameters such as number of nodes, number of edges, average node degree, average clustering coefficient and shortest path of the network, and identification of subnetworks and their novel interactions.

Close modal

Topological analysis of the network was performed for (a) node degree distribution P(k), (b) average clustering coefficient (<ck>), and (c) shortest path length (spl; Fig. 1). Node degree distribution measures the probability that a given protein interacts with k other proteins. As seen, network structure obeys power law P (k) ∼ kγ, in which γ is degree exponent and R2 value is computed to measure how well the data points fit to the curve, which taken together indicates that the distribution is scale free (γ = 1.322 and R2 = 0.852; Fig. 2A-I). Scale-free topology of any network implies robustness of the key seed proteins (25). Randomized networks on the contrary showed Poisson distribution of node degree and did not exhibit scale-free property (Fig. 2B-II). The clustering coefficient measures the tendency of proteins in a network to form the clusters (Fig. 2A-III). The average clustering coefficient of the nodes declines as the number of protein interactions per protein increases, indicating that these networks have a potential to form hierarchical organization. Average clustering coefficient for the randomized networks did not behave as real network (Fig. 2B-IV). Topological coefficient (Tn) is a relative measure of extent to which any protein in the network shares interaction partners with other proteins. Tn diminishes with the number of links (Fig. 2A-V), indicating that in this network, hubs do not have many common neighbors than proteins with fewer links. This is consistent with the property of modular network organization. Topological coefficient of randomized network did not behave similarly and was not modular (Fig. 2B-VI). The shortest path length of GBM network was lower compared with randomized networks, suggesting that signal transmission relay from one node to the other travels more rapidly (Fig. 2C).

Figure 2.

Various topological properties of GBM network are shown. A, I, node degree distribution plotted against number of nodes; II, distribution of average clustering coefficient in GBM network; III, topological coefficient plotted versus number of neighbors in GBM-specific network; IV, betweeness centrality is plotted against number of neighbors. B, I, distribution of node degree is plotted for randomized network; II, average clustering coefficient is plotted for respective randomized network; III, topological distribution for randomized network; IV, betweeness plotted for randomized network. C, comparative line graph of the shortest path length in real and random GBM network. D, comparison of subnetworks based on the Jaccard index.

Figure 2.

Various topological properties of GBM network are shown. A, I, node degree distribution plotted against number of nodes; II, distribution of average clustering coefficient in GBM network; III, topological coefficient plotted versus number of neighbors in GBM-specific network; IV, betweeness centrality is plotted against number of neighbors. B, I, distribution of node degree is plotted for randomized network; II, average clustering coefficient is plotted for respective randomized network; III, topological distribution for randomized network; IV, betweeness plotted for randomized network. C, comparative line graph of the shortest path length in real and random GBM network. D, comparison of subnetworks based on the Jaccard index.

Close modal

Next, we examined the features of the biological network that unites both local and global topological properties of the key connecting protein in the network in terms of its betweenness. Betweenness of a given node in a network is related to the number of times that node is a member of the set of shortest paths that connect all the pairs of nodes in the network. In biological terms, high betweenness of a node concordantly translates to a central regulatory attribute, which acts as an organizational functional module and that it can bring communication between distant proteins allowing one to relate local network structure to global network topology. Unlike the connectivity k that ranged from 1 to 362 in the interaction network, values for betweenness range over several orders of magnitude (Fig. 2A-VII). Majority of input seed proteins displayed high centrality values compared with their average value, implying their regulatory feature (Supplementary Table S3; ref. 20). Such nodes are absent in randomized networks (Fig. 2B-VIII).

Modular organization and correlation with molecular functions

The network generated using 62-seed proteins and their interactions was complex. To decipher the connecting molecules that had comprehensive effects, we dissected major network into subnetworks using topological filter option in Cytoscape. This filter allows in selecting nodes based on the properties of its nearby neighboring nodes. We used different topological cutoffs of ≥3, 4, and 5 to readily identify submodules that are interconnected through intermediary molecules and are involved in more than one biological process. This exercise resulted in 99 connecting molecules representing various combinations between hubs (Supplementary Table S4), which are interconnected through some key molecules.

To determine the functional coordination between the hubs, we calculated the Jaccard index for the three different subnetworks. The five component modules revealed high Jaccard indices, suggesting close functional collaboration compared with the three and four component subnetworks (Fig. 2D). Based on shared interactors between two proteins, we calculated the Z-score to determine the significance of the module in this analysis (Supplementary Table S4; ref. 21). The five component subnetworks showed higher Z-score value than other hubs. So we set a threshold value at ≥5 to identify highly interconnected subnetworks, which identified 15 connecting proteins between 22 highly linked hubs in the GBM network (Table 1). Recently, Yu and colleagues (26) have shown that all the hub proteins are characterized by high betweeness values also termed as bottleneck nodes. We have computed the betweeness values for five component subnetworks as described by Yu and colleagues (26), and all the hubs of the subnetwork showed high betweeness (Supplementary Table S5; Table 1). Of these, two highly connected subnetworks with CSK21 and PP1A as linker proteins for which no prior information on their involvement in glioma has been reported were identified. Interestingly, these two connecting molecules in addition to all other linking molecules of the 15 subnetworks showed highest betweenness values, suggesting that they are essential points (bottleneck nodes) in these subnetworks. To determine how compact and functionally relevant the CSK21 and PP1A modules are, we further performed statistical analysis separately on each of them. Both these subnetworks are connected with the shortest path length of <3, depicting high seed proximity. These subnetworks are explained in the following sections.

Table 1.

Significantly enriched modular subnetworks

Si.noSubnetworksConnecting moleculeZ-scoreBetweeness of Connecting moleculePubmed ID
CDC2-CAV1-RB-P53-PTEN PP1A 9.5 14,663.03 NA 
CDC2-PTEN-TOP2A-CAV1-P53 CSK21 8.82 12,593.93 NA 
CDC2-PTEN-MDM2-CCND1-CAV1-HMGB2-RB AR 21.67 22,583.28 9008101 
PTEN-CCND1-RB-MDM2-EGFR-CAV1-P53 ESR1 20.13 29,518.38 12794756 
MDM2-ASCL1-CCND1-CEBPD-RB-P53 EP300 10.18 10,623.24 14712485 
TIMP1-SERPINE1-THBS1-LTF-CAV1 LRP1 10.06 16,835.22 16230396 
EGFR-AEBP1-P53-CAV1-RB MK01 7.97 29,069.08 18410277 
CDC2-MDM2-EGFR-CAV1-RB-P53 ABL1 15.98 22,583.28 16507782 
CDC2-CCND1-P53-CCNB1-TOP2A PIN1 8.99 1,708.04 17938171 
10 IGFBP2-VEGFA-SERPINE1-THBS1-FN1 VTN 8.34 3,196.83 10389948 
11 PRC1-FABP5-CCND1-CCNB1-RB-P53 CDK2 9.4 23,650.72 19139420 
12 CCND1-CCNB1-RB-P53-CDC2 CDKN1A 10.14 2,882.68 15756520 
13 CDC2-CCND1-P53-CCNB1-TOP2A BRCA1 9.9 1,708.04 15367334 
14 CDC2-MDM2-CEBPD-RB-P53 E2F1 10.52 3,569.94 16264179 
15 MDM2-CCND1-TOP2A-RB-P53 HDAC1 10.53 3,043.91 18483381 
Si.noSubnetworksConnecting moleculeZ-scoreBetweeness of Connecting moleculePubmed ID
CDC2-CAV1-RB-P53-PTEN PP1A 9.5 14,663.03 NA 
CDC2-PTEN-TOP2A-CAV1-P53 CSK21 8.82 12,593.93 NA 
CDC2-PTEN-MDM2-CCND1-CAV1-HMGB2-RB AR 21.67 22,583.28 9008101 
PTEN-CCND1-RB-MDM2-EGFR-CAV1-P53 ESR1 20.13 29,518.38 12794756 
MDM2-ASCL1-CCND1-CEBPD-RB-P53 EP300 10.18 10,623.24 14712485 
TIMP1-SERPINE1-THBS1-LTF-CAV1 LRP1 10.06 16,835.22 16230396 
EGFR-AEBP1-P53-CAV1-RB MK01 7.97 29,069.08 18410277 
CDC2-MDM2-EGFR-CAV1-RB-P53 ABL1 15.98 22,583.28 16507782 
CDC2-CCND1-P53-CCNB1-TOP2A PIN1 8.99 1,708.04 17938171 
10 IGFBP2-VEGFA-SERPINE1-THBS1-FN1 VTN 8.34 3,196.83 10389948 
11 PRC1-FABP5-CCND1-CCNB1-RB-P53 CDK2 9.4 23,650.72 19139420 
12 CCND1-CCNB1-RB-P53-CDC2 CDKN1A 10.14 2,882.68 15756520 
13 CDC2-CCND1-P53-CCNB1-TOP2A BRCA1 9.9 1,708.04 15367334 
14 CDC2-MDM2-CEBPD-RB-P53 E2F1 10.52 3,569.94 16264179 
15 MDM2-CCND1-TOP2A-RB-P53 HDAC1 10.53 3,043.91 18483381 

NOTE: Z-Score, subnetwork significance; Pubmed ID, glioblastoma reference for connecting molecule; betweeness, a measure of total number of nonredundant shortest paths going through a certain node (see Supplementary Table S5).

CSK21 subnetwork module

This molecule centers around the subnetwork involving hubs associated with cell cycle such as P53, CDC2, PTEN, TOP2A, and CAV1 (Fig. 3A) consisting of 569 nodes (proteins) with a communication efficiency of 2.7 and a Z score of 8.82. CK2 or casein kinase 2 is a tetrameric protein formed by the association of homodimeric regulatory subunit β, and homodimeric or heterodimeric catalytic subunit α (CSK21) or α' (CSK22). CSK21/CK2 as a protein is a promiscuous serine/threonine kinase with over 300 targets defined thus far involved in a plethora of functions such as cell cycle regulation, transcriptional control, and apoptosis. Its subcellular localization also plays an important role in defining its function in cell survival. Besides being involved in cell proliferation and growth, its potent antiapoptotic activity supports cancer cell phenotype (27).

Figure 3.

A, subnetwork illustrates on CSK21 highlighted in dark orange color, which connects hubs P53, CDC2, PTEN, TOP2A, and CAV1. B, log2-transformed gene expression ratios obtained from real-time quantitative PCR analysis are plotted for CSK21. Dots, data derived from an individual patient sample. C, IHC staining pattern of CSK21 representative micrographs of (a) normal brain, (b) DA, and (c) AA tissue sections showing nuclear staining; (d) GBM tissue showing both cytoplasmic and nuclear staining; and (e) a negatively stained GBM tissue. All original magnifications are ×160.

Figure 3.

A, subnetwork illustrates on CSK21 highlighted in dark orange color, which connects hubs P53, CDC2, PTEN, TOP2A, and CAV1. B, log2-transformed gene expression ratios obtained from real-time quantitative PCR analysis are plotted for CSK21. Dots, data derived from an individual patient sample. C, IHC staining pattern of CSK21 representative micrographs of (a) normal brain, (b) DA, and (c) AA tissue sections showing nuclear staining; (d) GBM tissue showing both cytoplasmic and nuclear staining; and (e) a negatively stained GBM tissue. All original magnifications are ×160.

Close modal

Within this subnetwork, its role in cell cycle and association with p53 resulting in its phosphorylation and stabilization is well known in the regulation all the phases of mitosis. CDC2 phosphorylates CSK21 in a cell cycle–dependent manner during mitotic prophase and metaphase, failing which leads to mitotic catastrophe (28). CSK21 protein phosphorylates PTEN at COOH-terminus on Ser370 and Ser385, and affects its stability and regulates AKT pathway in a PTEN-dependent manner (29). CSK21 interacts with NH2-terminal domain and phosphorylates CAV1 (30). Increased levels of phosphorylated Cav1 have been associated with elevated GTP-RhoA levels in metastatic tumor cells of various tissue origins. It also stabilizes FAK1 association with focal adhesion and, hence, its turnover and in turn regulates tumor cell migration and invasion (31). In its role in transcriptional regulation, CSK21 is involved in a successful initiation by phosphorylation of several components of transcription machinery including Topoisomerase II α (27).

Evaluation of the functional enrichment of this subnetwork showed distribution of its component proteins across numerous significant biological pathways encompassing cell cycle, p53 signaling, apoptosis, and several oncogenic pathways (Supplementary Table S6). This led us to investigate mRNA levels of CSK21 in tumor samples, and this analysis showed a moderate level of upregulation in higher grades of astrocytoma with a P value of 0.0015 (Fig. 3B). We also analyzed the expression of CSK21 at protein level in different grades of GBM by IHC (Fig. 3C). The mean ± SD LI for CSK21 nuclear expression in GBM was 14.23 ± 14.979 compared with AA (10.000 ± 10.607), DA (13.75 ± 12.500), and control (25.00 ± 0.000). Although astrocytomas of all grades of malignancy showed clear decrease in the nuclear protein expression with respect to control samples, the difference in the labeling between groups (DA, AA, and GBM) was not statistically significant. On the other hand, an increased cytoplasmic expression of CSK21 was evident with increasing grades of malignancy within astrocytoma grades (control, 0.000 ± 0.000; DA, 3.75 ± 7.500; AA, 4.00 ± 8.944; GBM, 10.00 ± 9.789), although not reaching statistical significance (Table 2).

PP1A subnetwork module

Table 2.

IHC LI of CSK21 and PP1A subcellular localization

Control brainDA (grade II)AA (grade III)GBM (grade IV)Post hoc PKruskal Wallis test
χ2P
CSK21 nuclear 25.00 ± 0.000 13.75 ± 12.500 10.00 ± 10.607 14.23 ± 14.979 ¥ = 0.584 3.143; 0.370 
Φ = 1.000 
graphic
= 0.927 
CSK21 cytoplasm 0.00 ± 0.000 3.75 ± 7.500 4.00 ± 8.944 10.00 ± 9.789 ¥ = 0.316 5.080; 0.166 
Φ = 0.611 
graphic
= 0.579 
PP1a nuclear 0.00 ± 0.000 0.00 ± 0.000 0.00 ± 0.000 12.31 ± 8.807 ¥ = 0.041* 13.662; 0.003 
Φ = 0.019* 
graphic
= 0.010* 
PP1a cytoplasm 0.00 ± 0.000 3.75 ± 4.787 1.00 ± 2.236 13.08 ± 10.112 
graphic
= 0.040* 
10.008; 0.018 
¥ = 0.076 
Φ = 0.198 
Control brainDA (grade II)AA (grade III)GBM (grade IV)Post hoc PKruskal Wallis test
χ2P
CSK21 nuclear 25.00 ± 0.000 13.75 ± 12.500 10.00 ± 10.607 14.23 ± 14.979 ¥ = 0.584 3.143; 0.370 
Φ = 1.000 
graphic
= 0.927 
CSK21 cytoplasm 0.00 ± 0.000 3.75 ± 7.500 4.00 ± 8.944 10.00 ± 9.789 ¥ = 0.316 5.080; 0.166 
Φ = 0.611 
graphic
= 0.579 
PP1a nuclear 0.00 ± 0.000 0.00 ± 0.000 0.00 ± 0.000 12.31 ± 8.807 ¥ = 0.041* 13.662; 0.003 
Φ = 0.019* 
graphic
= 0.010* 
PP1a cytoplasm 0.00 ± 0.000 3.75 ± 4.787 1.00 ± 2.236 13.08 ± 10.112 
graphic
= 0.040* 
10.008; 0.018 
¥ = 0.076 
Φ = 0.198 

NOTE: ¥, normal vs GBM; Φ, DA vs GBM;

graphic
, AA vs GBM.

*Tukey's post hoc test showing a significant P value in bold.

This subnetwork consists of 661 nodes with an average path length 2.9; Z-score of 9.5; centered around PP1A, integrating PTEN and CAV1, along with P53, CDC2, and retinoblastoma (RB; Fig. 4A). PP1A is a multifunctional protein phosphatase (Ser/Thr) that is involved in regulating various cellular processes (32). Important signal transducers such as PKA, AKT, PKC, cAMP-responsive element binding protein, GSK3β, Wee1, APC, and mitogen-activated protein kinase are the substrates for PP1A-mediated dephosphorylation (33). Association of PP1A with each of the main nodes in the module is described below.

Figure 4.

A, submodule showing PP1A as intermediate connecting link highlighted in dark orange color between P53, CDC2, PTEN, CAV1, and Rb. B, log2-transformed gene expression ratios obtained from real-time quantitative PCR analysis are plotted for PP1A. Dots, data derived from an individual patient sample. C, IHC staining pattern of PP1a representative micrographs of (a) normal brain, (b) DA, and (c), AA tissue sections showing absence of nuclear and cytoplasmic staining; and (d) GBM tissues showing cytoplasmic and (e) both cytoplasmic and nuclear staining, respectively. All original magnifications are ×160 except in d and e, which is ×320.

Figure 4.

A, submodule showing PP1A as intermediate connecting link highlighted in dark orange color between P53, CDC2, PTEN, CAV1, and Rb. B, log2-transformed gene expression ratios obtained from real-time quantitative PCR analysis are plotted for PP1A. Dots, data derived from an individual patient sample. C, IHC staining pattern of PP1a representative micrographs of (a) normal brain, (b) DA, and (c), AA tissue sections showing absence of nuclear and cytoplasmic staining; and (d) GBM tissues showing cytoplasmic and (e) both cytoplasmic and nuclear staining, respectively. All original magnifications are ×160 except in d and e, which is ×320.

Close modal

Phosphorylation of the catalytic subunit of PP1A by cdc2-cyclin B at Thr 320 regulates its activity in a cell cycle–dependent manner, whereas its dephosphorylation blocks cell cycle progression. This also mandates the destruction of cyclin B, Cdc2 kinase inactivation, and dephosphorylation of mitotic phosphoproteins. At the turn of the M-G1 phase, several mitotic proteins should be dephosphorylated to promote the timely dephosphorylation of mitotic substrates (34). PP1A phosphatase activity is regulated by its association with inhibitor proteins 1 and 2. PP1A is also phosphorylated by cdc2/LaminB complex. In addition, PKA-mediated phosphorylation of inhibitor 1 associates and keeps PP1A in an inactive form. For the next phase of the cell cycle to ensue, it is important that inactivated phosphorylated form of PP1A is reversed. This reversal happens in two stages. As cyclin B is destroyed at mitotic exit, cdc2 kinase activity drops, allowing autodephosphorylation of PP1A resulting in its partial activation. This is followed by PP1A-mediated dephosphorylation of Inh1 and its dissociation, allowing full activation of PP1A that can then dephosphorylate other mitotic proteins to allow mitotic exit (34). Another late-mitotic substrate of PP1A is Rb, which is hyperphosphorylated in the S phase until the end of mitosis. Hypophosphorylated form of Rb in G phase sequesters E2F, inhibiting its transcriptional activity needed for cells to exit from G1 and to enter S phase (35). Its association with p53 is shown by the observation that downregulation of PP1A impairs the p53-dependent induction of p21 upon DNA damage in inducing growth arrest and senescence (36). Coincidently, catalytic subunit of PP1A is shown to colocalize with pRb during senescence (37). CAV1 inhibits the dephosphorylation activities of PP1A through interaction with scaffolding domain-binding site resulting in the maintenance of phosphorylated Akt. This further describes the dysregulation of downstream signaling governed through the phosphorylation status of PP1A substrates such as GSK3β, FKHRLI, MDM2, BAD, IKK α, and β, ensuing Cav1-mediated cell survival (38). Interaction of PP1A with PTEN was reported in an antibody array screen to identify novel PP1A-interacting proteins (39), the functional relevance of which is still to be determined.

Gene Ontology enrichment analysis of the proteins of this module centered around PP1A, shows predominant association with cell cycle, P53 signaling, transforming growth factor-β signaling, adhesion, vascular endothelial growth factor (VEGF) signaling, apoptosis, regulation of actin cytoskeleton, ErbB signaling, neurodegenerative diseases, and several carcinogenic pathways with highly significant P values (Supplementary Table S7).

Next, we assessed the expression of PP1A at the mRNA level by real-time quantitative PCR in glial tumors. Overexpression of PP1A mRNA was seen in a grade-dependent manner in diffusely infiltrating astrocytomas (Fig. 4B). The transcript levels of PP1A were upregulated >20-fold (log2 ratio of 5) in a majority of GBMs (93.20%; 41 of 43). The difference in the transcript levels of GBM when compared with AA (11.54%; 3 of 26), DA (0.00%; 0 of 8), and control (12.5%; 1 of 8) was highly significant (P < 0.0001). Analysis of PP1A expression at the protein level by IHC revealed that there was no nuclear expression in AA, DA, and control samples (Fig. 4C). Nuclear expression of PP1A protein was seen only in GBM samples. The mean ± SD LI for PP1a nuclear expression in GBM was 12.31 ± 8.80 compared with AA, DA, and control (0.0000 ± 0.0000). The difference between GBM and control (P = 0.041), GBM and DA (P = 0.019), GBM and AA was statistically significant (P = 0.01). For PP1a cytoplasmic expression, the mean ± SD LI in GBM was 13.08 ± 10.11, compared with AA (1.00 ± 2.23), DA (3.75 ± 4.78), and control (0.0000 ± 0.0000). Among the groups, the difference between GBM and AA was significant (P = 0.04; Table 2).

A systems level approach to elucidate highly connected pathway networks is being increasingly exploited to understand the biology of cellular phenotype in disease conditions. In this study, we have used systems approach to identify key connecting molecules of important subnetworks from a GBM-specific network constructed using upregulated genes. The constructed GBM network revealed scale-free architecture compared with randomized network. The hub genes were highly connected and communicated with very high efficiency as evident by the shortest path length. The average clustering coefficient of the GBM network showed >5-fold increase compared with randomized network. Thus, the network created was compact and modular. Use of different topological cutoffs identified 99 subnetworks of degree 3, 4, and >5 component modules each with a unique connecting molecule. These hubs also showed high Jaccard indices relating to their close functional involvement. Two modules that stood out novel in this study, centered on CSK21 and PP1A in this study. Both these modules orchestrate many of the genes participating in cell cycle function and regulation. Further data mining revealed significant involvement of the subnetwork genes as identified through gene ontology analysis encompassing various tumor developmental pathways.

CK2 protein functions in a wide range of biological processes such as cell cycle, apoptosis, and transcriptional control. Its association with various hub molecules suggests diversity in the biological pathways in which it participates (27). CK2 has been implicated in the G2-M transition through its association with the mitotic spindle, centrosomes, as well as cell cycle regulatory proteins, such as p34cdc2, cdc34, and topoisomerase II (40). As evident from the immunohistochemical analysis of CSK21, there was overexpression of CSK21 in the cytoplasm of tumor cells with increasing grades of malignancy along with an associated decrease in nuclear expression, indicating that cytoplasmic CSK21 might be closely associated with the aggressive form of glioma tumors. However, it must be mentioned here that the expression pattern of CSK21 was not uniform in all GBM cases, and four cases of GBM did not show the expression of CSK21 at all. It is quite likely that this might be due to the existence of specific subclasses with in GBM that are being reported recently (12). Owing to the multifunctionality of this protein, it is difficult to predict at this juncture the underlying functions of CSK21 that is important in driving oncogenesis. It is noteworthy that CSK21 has recently emerged as potential therapeutic target in various proliferation abnormalities (40).

The other highly expressed intermediate molecule PP1A that we have identified in the present study is a serine threonine protein phosphatase with several substrates important in proper functioning of the cell cycle, metabolism, actomyosin reorganization, and apoptosis (32). Members of the PPP family are highly expressed in neurons controlling signaling matrix. Intricate interplay of dephosphorylation and kinase activity governs the normal process of cell cycle. Genetic alterations in the catalytic and regulatory subunits of these protein phosphatases have been reported in various human cancers including oral squamous cell carcinoma and prostate cancer (41, 42). In an extensive study of genetic alterations compiled using oligonucleotide-based array, comparative genomic hybridization showed that 11q13.4 is differentially deleted specifically in primary glioma as opposed to secondary GBM and has been classified as high-priority primary GBM subtype–unique minimal common regions. This deleted region however does not harbor PP1A (43). Our present report is the first demonstration of a highly statistically significant upregulation of PP1A both at the mRNA and protein level in GBM. The high nuclear and cytoplasmic expression of PP1A in GBM when compared with the other grades suggests that this molecule might play a very important role in the malignant progression of glioma. It is also interesting to note that synthetic and natural inhibitors of PP1A have also been identified (44), which are being tried for possible therapeutic intervention in other types of cancer. These inhibitory molecules may also prove useful in clinical management of GBM. In summary, this study using systems biology approach has identified two proteins CSK21 and PP1A that connect important hubs, one being a kinase and the other being a phosphatase, which might modulate the key steps of cell cycle–promoting gliomagenesis. These proteins are deregulated in other cancers as well. Mechanistically, these proteins may function in variety of ways within cellular environment because they have a large number of interactors. Their upstream and downstream regulation at functional level will be crucial to get insight in their roles, in addition to their posttranslational modification status. A more comprehensive study on the tumorogenic role of PP1A and CSK21 with their associated proteins will foster our understanding of the role of these proteins in glioma progression.

No potential conflicts of interest were disclosed.

Grant Support: NMITLI program of Council of Scientific and Industrial Research and Department of Biotechnology, New Delhi. M.R.S. Rao is a JC Bose Fellow of Department of Science and Technology.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1
Kleihues
P
,
Louis
DN
,
Wiestler
OD
,
Burger
PC
,
Schiethauer
BW
. 
WHO grading of tumors of central nervous system
. In:
Louis
DN
,
Ohgaki
H
,
Wiestler
OD
,
Cavenee
WK
, editors.
WHO classification of tumors of the central nervous system
. 4th ed
Lyon (France)
:
IARC Press
; 
2007
, p.
10
1
.
2
Kleihues
P
,
Ohgaki
H
. 
Primary and secondary glioblastomas: from concept to clinical diagnosis
.
Neuro-oncol
1999
;
1
:
44
51
.
3
Ruano
Y
,
Ribalta
T
,
de Lope
AR
, et al
. 
Worse outcome in primary glioblastoma multiforme with concurrent epidermal growth factor receptor and p53 alteration
.
Am J Clin Pathol
2009
;
131
:
257
63
.
4
Rhodes
DR
,
Kalyana-Sundaram
S
,
Mahavisno
V
,
Barrette
TR
,
Ghosh
D
,
Chinnaiyan
AM
. 
Mining for regulatory programs in the cancer transcriptome
.
Nat Genet
2005
;
37
:
579
83
.
5
Kim
WJ
,
Bae
SC
. 
Molecular biomarkers in urothelial bladder cancer
.
Cancer Sci
2008
;
99
:
646
52
.
6
Martinez-Glez
V
,
Franco-Hernandez
C
,
Rey
JA
. 
Microarray gene expression profiling in meningiomas and schwannomas
.
Curr Med Chem
2008
;
15
:
826
33
.
7
Somasundaram
K
,
Reddy
SP
,
Vinnakota
K
, et al
. 
Upregulation of ASCL1 and inhibition of Notch signaling pathway characterize progressive astrocytomas
.
Oncogene
2005
;
24
:
7073
83
.
8
Zhou
H
,
Miki
R
,
Eeva
M
, et al
. 
Reciprocal regulation of SOCS 1 and SOCS3 enhances resistance to ionizing radiation in Glioblastoma
.
Clin Cancer Res
2007
;
13
:
2344
53
.
9
Reddy
SP
,
Britto
R
,
Vinnakota
K
, et al
. 
Novel glioblastoma markers with diagnostic and prognostic value identified through transcriptome analysis
.
Clin Cancer Res
2008
;
14
:
2978
87
.
10
Reddy
SP
,
Britto
R
,
Vinnakota
K
, et al
. 
PBEF1/NAmPRTase/Visfatin: a postential malignant astrocytomas/glioblastoma serum marker with prognostic value
.
Cancer Biol Ther
2008
;
7
:
663
8
.
11
Phillips
HS
,
Kharbanda
S
,
Chen
R
, et al
. 
Molecular subclasses of high grade glioma predict prognosis, delineate a pattern of disease progression and resemble stages in neurogenesis
.
Cancer Cell
2006
;
9
:
157
73
.
12
Verhaak
RG
,
Hoadley
KA
,
Purdom
E
, et al
. 
Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR and NF1
.
Cancer Cell
2010
;
17
:
98
110
.
13
Chuang
HY
,
Lee
E
,
Liu
YT
,
Lee
D
,
Ideker
T
. 
Network-based classification of breast cancer metastasis
.
Mol Syst Biol
2007
;
3
:
140
.
14
Horvath
S
,
Zhang
B
,
Carlson
M
, et al
. 
Analysis of oncogenic signaling networks in glioblastoma identifies ASPM as a molecular target
.
Proc Natl Acad Sci U S A
2006
;
103
:
17402
7
.
15
Sallinen
SL
,
Sallinen
PK
,
Haapasalo
HK
, et al
. 
Identification of differentially expressed genes in human gliomas by DNA microarray and tissue chip techniques
.
Cancer Res
2000
;
60
:
6617
22
.
16
Khatua
S
,
Peterson
KM
,
Brown
KM
, et al
. 
Over expression of the EGFR/FKBP12/HIF-2α pathway identified in childhood astrocytomas by angiogenesis gene profiling
.
Cancer Res
2003
;
63
:
1865
70
.
17
Hernandez-Toro
J
,
Prieto
C
,
De las Rivas
J
. 
APID2NET: unified interactome graphic analyzer
.
Bioinformatics
2007
;
23
:
2495
7
.
18
Shannon
P
,
Markiel
A
,
Ozier
O
, et al
. 
Cytoscape: a software environment for integrated models of biomolecular networks
.
Genome Res
2003
;
13
:
2498
504
.
19
Assenov
Y
,
Ramirez
F
,
Schelhorn
SE
,
Lengauer
T
,
Albrecht
M
. 
Computing topological parameters of biological networks
.
Bioinformatics
2008
;
24
:
282
4
.
20
Scardoni
G
,
Petterlini
M
,
Laudanna
C
. 
Analyzing biological network parameters with CentiScape
.
Bioinformatics
2009
;
25
:
2857
9
.
21
Gong
Y
,
Kakihara
Y
,
Krogan
N
, et al
. 
An atlas of chaperone- protein interactions in sacchromyces cerevisiae: implications to protein folding pathways in the cell
.
Mol Syst Biol
2009
;
5
:
275
.
22
Santosh
V
,
Arivazhagan
A
,
Reddy
P
, et al
. 
Grade specific expression of IGFBP-2, -3 and -5 in astrocytomas: IGFBP3 emerges as a strong predictor of survival in patients with newly diagnosed glioblastoma
.
Cancer Epidemiol Biomarkers Prev
2010
;
19
:
1399
408
.
23
Corder
GW
,
Foreman
DI
. 
Nonparametric statistics for non-statisticians: a step-by-step approach
. In:
Corder
GW
,
Foreman
DI
, editors.
Comparing more than unrelated samples: the Kruskal Wallis H- Test
.
New Jersey
:
Wiley
; 
2009
, p.
99
118
.
24
Kruskal
WH
,
Wallis
WA
. 
Use of ranks in one-criterion variance analysis
.
J Am Stat Assoc
1952
;
47
:
583
621
.
25
Barabasi
AL
,
Oltvai
ZN
. 
Network biology: understanding the cell's functional organization
.
Nat Rev Genet
2004
;
5
:
101
13
.
26
Yu
H
,
Kim
PM
,
Sprecher
E
,
Trifonov
V
,
Gertstein
M
. 
The importance of bottlenecks in protein networks: correlation with gene essentiality and expression dynamics
.
PLOS Comput Biol
2007
;
3
:
e59
.
27
Trembley
JH
,
Wang
G
,
Unger Slaton
J
,
Ahmed
K
. 
Protein Kinase CK2 in health and disease: CK2: a key player in cancer biology
.
Cell Mol Life Sci
2009
;
66
:
1858
67
.
28
St-Denis
NA
,
Litchfield
DW
. 
Protein kinase CK2 in health and disease: from birth to death: the role of protein kinase CK2 in the regulation of cell proliferation and survival
.
Cell Mol Life Sci
2009
;
66
:
1817
29
.
29
Miller
SJ
,
Lou
DY
,
Seldin
DC
,
Lane
WS
,
Neel
BG
. 
Direct identification of PTEN phosphorylation
.
FEBS Lett
2002
;
528
:
145
53
.
30
Sargiacomo
M
,
Scherer
PE
,
Tang
ZL
,
Casanova
JE
,
Lisanti
MP
. 
In vitro phosphorylation of caveolin-rich membrane domains: identification of an associated serine kinase activity as a casein kinase II-like enzyme
.
Oncogene
1994
;
9
:
2589
95
.
31
Joshi
B
,
Strugnell
SS
,
Goetz
JG
, et al
. 
Phosphorylated caveolin-1 regulates Rho/ROCK-dependent focal adhesion dynamics and tumor cell migration and invasion
.
Cancer Res
2008
;
68
:
8210
20
.
32
Ceulemans
H
,
Bollen
M
. 
Functional diversity of protein phosphatase-1 a cellular economizer and reset button
.
Physiol Rev
2004
;
84
:
1
39
.
33
Cohen
PT
. 
Protein phosphatase1-targeted in many directions
.
J Cell Sci
2002
;
115
:
241
56
.
34
Wu
JQ
,
Guo
JY
,
Tang
W
, et al
. 
PP1-mediated dephophorylation of phosphoproteins at mitotic exit is controlled by inhibitor-1 and PP1 phosphorylation
.
Nat Cell Biol
2009
;
11
:
644
51
.
35
Ludlow
JW
,
Nelson
DA
. 
Control and activity of type-1 serine/threonine protein phosphatase during cell cycle
.
Semin Cancer Biol
1995
;
6
:
195
202
.
36
Castro
ME
,
Ferrer
I
,
Cascón
A
, et al
. 
PPP1CA contributes to the senescence program induced by oncogenic Ras
.
Carcinogenesis
2008
;
29
:
491
9
.
37
Ruiz
L
,
Traskine
M
,
Ferrer
I
, et al
. 
Characterization of the p53 response to oncogene-induced senescence
.
PLos One
2008
;
3
:
e3230
.
38
Li
L
,
Ren
CH
,
Tahir
SA
,
Ren
C
,
Thompson
TC
. 
Caveolin-1 maintains activated Akt in prostate cancer cells through scaffolding domain binding site interactions with and inhibition of serine/threonine protein phosphatases PP1 and PP2A
.
Mol Cell Biol
2003
;
23
:
9389
404
.
39
Flores-Delgado
G
,
Liu
CW
,
Sposto
R
,
Berndt
N
. 
A limited screen for protein interactions reveals new roles for protein phosphatase 1 in cell cycle control and apoptosis
.
J Proteome Res
2007
;
6
:
1165
75
.
40
Duncan
JS
,
Litchfield
DW
. 
Too much of as good thing: the role of protein kinase CK2 in tumorigenesis and prospects for therapeutic inhibition of CK2
.
Biochim Biophys Acta
2008
;
1784
:
33
47
.
41
Hsu
LC
,
Huang
X
,
Seasholtz
S
,
Potter
DM
,
Gollin
SM
. 
Gene amplification and overexpression of protein phosphatase 1α in oral squamous cell carcinoma cell lines
.
Oncogene
2006
;
25
:
5517
26
.
42
Prowatke
I
,
Devens
F
,
Benner
A
, et al
. 
Expression analysis of imbalanced genes in prostate carcinoma using tissue microarrays
.
Br J Cancer
2007
;
96
:
82
8
.
43
Maher
EA
,
Brennan
C
,
Wen
PY
, et al
. 
Marked genomic differences characterize primary and secondary glioblastoma subtypes and identify two distinct molecular and clinical secondary glioblastoma entities
.
Cancer Res
2006
;
66
:
11502
13
.
44
McCluskey
A
,
Sim
AT
,
Sakoff
JA
. 
Serine-threonine protein phosphatase inhibitors: development of potential therapeutic strategies
.
J Med Chem
2002
;
14
:
1151
75
.