Abstract
Background
To investigate differential egonetwork modules and pathways in glioma using EgoNet algorithm.
Methodology
Based on microarray data, EgoNet algorithm mainly comprised three stages: construction of differential co-expression network (DCN); EgoNet algorithm used to identify candidate ego-network modules based on the increased classification accuracy; statistical significance for candidate modules using random permutation testing. After that, pathway enrichment analysis for differential ego-network modules was implemented to illuminate the biological processes.
Results
We obtained 109 ego genes. From every ego gene, we progressively grew the ego-networks by levels; we extracted 109 ego-networks and the mean node size in an ego-network was 6. By setting the classification accuracy threshold at 0.90 and the count of nodes in an ego-network module at 10, we extracted 8 candidate ego-network modules. After random permutation test with 1000 times, 5 modules including module 59, 72, 78, 86, and 90 were identified to be significant. Of note, the genes of module 90 and 86 were enriched in the pathway of resolution of sister chromatid cohesion and mitotic prometaphase, respectively.
Conclusion
The identified modules and their corresponding ego genes might be beneficial in revealing the pathology underlying glioma and give insight for future research of glioma.
1 Introduction
Glioma is the most common primary tumor in the brain [1]. According to their cellular origins, World Health Organization system classifies glioma into oligodendrogliomas, astrocytomas, ependymomas and mixed tumors [2]. The main symptom of glioma patients is increased intracranial pressure, which gives rise to headache, paropsia, nausea and vomiting. Although great improvements have been made in diagnosis and therapy of glioma, this disease invariably results in death within months or years because of the high invasiveness of cancer cells [3]. Glioma has been a major health problem in the world, but the knowledge about the pathogenesis underlying glioma is limited. Thus, a better understanding of the molecular mechanisms of this disease is helpful to develop the novel therapeutic methods for glioma.
Recently, with the development of biotechnology and the innovation of high throughput technique, researchers have focused on investigating the occurrence and progression of diseases at the genome level. Numerous studies have identified biomarkers associated with diseases by gene expression profiling [4, 5]. To a certain degree, these biomarkers offer a valuable platform for therapeutic strategy. However, signatures based on gene expression data alone are not reliable [6]. To surmount this difficulty, network biology has rapidly evolved as a promising area which is sufficient to elucidate molecular mechanisms, because it addresses the intrinsic structure and organization of networks of biological interactions, rather than the individual gene [7, 8]. In addition, the extent of network size may result in overlooking a certain amount of important genes and interactions [9]. Thus, identifying modules or sub-networks of the intricate network can avoid this kind of difficulty [10, 11]. So far, several methods have been created to integrate microarray profiles with protein-protein interaction (PPI) maps or pathway databases, with the goal of extracting significant module markers for predicting biological or clinical outcomes [12-14]. Of note, these methods are greatly heuristic, and the definition of output sub-networks or modules is indeterminate without formal topological characteristics. Fortunately, a method called EgoNet [15] can identify significant modules that are functionally related to diseases, and the type of modules selected using EgoNet are defined as ego-network modules which are well defined in the social networks study [16].
To classify glioma samples, we identified differential ego-network modules by integrating the gene expression profiles and the human PPI network through a module based method of EgoNet. First, a dataset obtained from the Gene Expression Omnibus (GEO) database (GSE15824) was downloaded for subsequent analysis. Afterwards, a differential co-expression network (DCN) was constructed, followed by the selection of ego genes. Starting with these ego genes, all ego-network modules were collected based on the increased classification accuracy via snowball sampling procedure. Subsequently, random permutation testing was used to evaluate the significance of the identified modules. Finally, pathway enrichment analysis for differential modules was implemented to illuminate the biological processes. These identified modules and pathways might provide novel avenues for more advanced investigation into the pathological mechanism of glioma. Improving the understanding of the pathogenic processes of glioma may provide evidence that aids in the development of early diagnosis and treatment strategies for glioma patients.
2 Materials and methods
2.1 Data resources and pre-processing
The microarray profile of glioma, with accession number GSE15824, was downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/), which was deposited in the GPL570 platform of Affymetrix Human Genome U133 Plus 2.0 Array. A total of 45 samples were available for subsequent analysis, including 40 glioma samples and 5 normal tissue samples. The annotation data of all probes were supplied by Affymetrix where the raw data file was downloaded [17].
The raw probe data in CEL files were transformed into the recognizable expression measures [18], in which the probe expressions were standardized using robust multi-array average function. Then, the probes were converted into the corresponding gene symbols via the R/Bioconductor annotation package as well as the Affymetrix Human Gene 2.0 ST Array. Significantly, probes without gene symbols were dismissed. For each sample, the expression values of all probes for a given gene were reduced to a single value by taking the average expression value [19]. After mapping the probes to the gene annotations, we obtained 17351 genes in total.
2.2 PPI network construction
The PPI analysis is a research tool utilized to uncover the functions of proteins which can help to elucidate cellular activities such as growth, development, differentiation, metabolism, and apoptosis at the molecular level. Moreover, many function-associated genes can co-express and the gene expression levels are varied in different cell type and state. Therefore, an original PPI network involved in 787896 interactions and 16730 genes was obtained from the database of Search Tool for the Retrieval of Interacting Genes (STRING) (http://string-db.org/) [20]. Then, taking the intersection of the interactions of the original PPI network with the microarray data, we obtained a new PPI network, which included 46460 interactions (7561 genes).
2.3 Construction of DCN
The DCN was constructed in the following procedure. First of all, we picked out the edges from then new PPI network to construct a binary co-expression network according to the absolute value of Pearson correlation coefficient (PCC) of interactions of the new PPI network. In our study, only edges whose PCC values were higher than the cutoff criteria δ (δ = 0.8) were selected. Then, one-side t-test was employed to identify differential gene expression, and the edges of the binary co-expression network were assigned the weight values based on the P values of differential expression between glioma and control groups.
2.4 Assumption of the novel method of EgoNet
An ego-network module is a part of a network that refers to a particular node we are focusing on, which we call an ego gene. In addition to the ego gene, the network also involves a neighborhood containing all nodes to which the ego is connected to at a certain path length. The onestep neighborhood includes nodes that the ego gene is directly connected to (referred to as the alters of ego).
The assumption of this method is that if most neighbors of a central disease gene are disease genes, then other neighboring genes may participate in the disease pathway (see Figure 1A). Alternatively, if the majority of neighbors of the ego gene are related to a disease, the ego gene itself is regarded highly likely to exert a function in this disease, as shown in Figure 1B. The goal of this method is to uncover the hidden genes which exhibit no significance by themselves, yet are clustered in a module. This algorithm took the PPI network and gene expression profile as input, then EgoNet repeatedly scanned through all genes with two or more neighbors in the network. Taking every ego gene as each initial node, the score of the level-one ego-network was calculated based on how well the genes as a collection predicted the classification accuracy. Subsequently, it spread outward from the ego gene progressively to include more genes in the predictive model. The process discontinued when the classification accuracy decreased. The procedure of growing egonetwork modules was defined as snowball sampling. After obtaining the score of a module, the significance was assessed by permutation testing to evaluate the significance of the modules. This algorithm mainly included three steps: ego genes identification, snowball sampling utilized to extract ego-network modules, and refinement of ego-network modules.
Two illustrative ego-networks. Yellow nodes were on behalf of putative disease genes, white nodes stood for hidden disease genes either as alter nodes (A) or ego node (B).
2.4.1 Ego genes identification
In this step, genes in the DCN were sorted based on the topological characteristics of the genes in the DCN. In short, we constructed a function h:V→R to evaluate the importance of node i in the DCN.
and A′ijk = D−1/2AijkD1/2
In this function, h(i) stood for the importance of node i; Nk(i) was the set of neighbors of gene i in the DCN; A′ijk stood for the degree normalized weighted adjacency matrix and D was diagonal matrix.
For each gene, after acquiring its ranks in DCN, marked as h = [h(1),…, h(M)], a z-score was computed for each rank h(l). Subsequently, the rank was obtained for that gene in the DCN via averaging the z-scores. The top 5% genes in the DCN were selected and defined as ego genes.
2.4.2 Ego-network module search based on classification accuracy using snowball sampling
Starting with the ego genes obtained above, the egonetwork module search procedure exhaustively included genes whose addition caused the maximum increase in the classification accuracy until there was no additional increase. This process of spreading ego-network modules is known as snowball sampling [21].
For a given ego gene u∊U, we determined an egonetwork module C={u}. Afterwards, the gene v which was the neighbor of gene u was successively added into the module C and a new module C’ was obtained. Then, we calculated the change of classification accuracy ΔS, that was to say, ΔS (C’, C) = S(C) − S(C’). When ΔS (C’, C) was greater than 0, this indicated that the addition of gene v increased the classification accuracy of the ego-network module C. Thus, gene v would be added to module C. If there were more than one node contained at each stage, we randomly selected one. Similarly, all the neighbors with ΔS > 0 were added into the module C when the classification accuracy dropped.
2.4.3 Refinement of ego-network modules
In this stage, we abandoned several ego-network modules whose size was smaller than 10. Moreover, modules with classification accuracy less than 0.90 were also eliminated.
2.5 Calculation of the statistical significance of candidate modules
The statistical significance of ego-network modules was calculated based on the null score distribution of classification accuracy generated by random permutation test. To derive the scores for the null distribution, we implemented random permutation test 1000 times on the same module. Based on the null distribution, the P value of an ego-network module was calculated as the probability of the candidate ego-network module having the smaller accuracy value using the following formula:
In this formula, N(HR) stood for the count of random permutation test, S(HR) was the accuracy value of egonetwork modules generated by the random permutation test. S(HC) represented the classification accuracy of candidate ego-networks.
Then, P values were adjusted for multiple testing via the Benjamini & Hochberg method [22]. In our study, a false discovery rate (FDR) of 0.01 was employed as the significance threshold.
2.6 Pathway enrichment analysis
For preliminary investigation into the functional differences of genes in differential ego-network modules in the glioma and control group, we performed the pathway enrichment analysis for the genes of differential modules. Briefly, all human reference pathways were downloaded from Reactome database (http://www.reactome.org). Pathways with too few genes might not have enough biological information, and pathways with too many genes might be too generic [23]. Thus, the intersection of the genes of each reference pathway with the genes in the new PPI network was extracted. After excluding the pathways with less than 5 genes or more than 100 genes, a set of informative pathways was obtained for subsequent analysis. Next, genes in differential ego-network modules were mapped to each reference pathway so that we derived the pathways enriched by each differential ego-network module. Then, Fisher’s exact test was utilized to assess the enrichment effects. After that, the Benjamini & Hochberg method [22] was employed to compute FDR to further correct the P values. In the current study, pathways with FDR < 0.05 were regarded as the pathways enriched by a given ego-network module. Significantly, the pathways enriched by each module were sorted in ascending order based on FDR values, and the pathway with the smallest FDR was extracted as the significant pathway of a given ego-network module.
2.7 Statistical analysis
The Feature Extraction software 10.7 was used to analyze the statistical significance of the microarray results. Quantile Algorithm (Agilent Technologies) was applied to normalize the raw data. The FDR was calculated to correct the raw P-values. The cut-off criteria employed to designate differential modules was a FDR < 0.01, as well as the threshold for significant pathways was a FDR < 0.01. P < 0.05 was considered to indicate a statistically-significant difference. Fisher’s exact test was used to select the significant pathways. In the current study, SPSS version 18.0 (SPSS Inc., Chicago, IL, USA) was used for statistical analysis.
Ethical approval
The conducted research is not related to either human or animals use
3 Results
3.1 Ego-network modules identification
A DCN including 6325 edges among 2184 nodes was constructed. Based on the distribution of z-scores of genes in DCN, a total of 109 genes were identified as ego genes. The top 20 ego genes are depicted in Table 1. Subsequently, starting with every ego gene, we progressively grew the ego-networks by levels. From every ego-network, this process ceased if the classification accuracy values dropped along with the growth. Finally, a total of 109 ego-network modules were identified and the mean size of nodes in an ego-network was 6. By setting the classification accuracy threshold at 0.90 and the count of nodes in an ego-network module at 10, we screened out 8 candidate ego-network modules, including module 59 (accuracy = 1.00 and count = 10), module 72 (accuracy = 0.90 and count = 12), module 78 (accuracy = 1.00 and count = 11), module 4 (accuracy = 0.92 and count = 12), module 62 (accuracy = 0.92 and count = 10), module 79 (accuracy = 0.90 and count = 10), module 86 (accuracy = 1.00 and count = 12), and module 90 (accuracy = 0.96 and count = 14).
The top 20 ego genes identified in the differential co-expression network (DCN) and the distribution of their z-scores
| Ego gene | Z-score | Ego gene | Z-score |
|---|---|---|---|
| MCHR1 | 200.5926 | CDC6 | 143.2547 |
| SSTR4 | 180.2623 | TTK | 142.3731 |
| PLCB2 | 179.1958 | TRIP13 | 141.2547 |
| HRH4 | 168.02118 | MTNR1B | 140.9077 |
| MAPK1 | 160.79336 | GRP | 140.4263 |
| NMUR2 | 155.6705 | NUSAP1 | 138.9907 |
| CCNB1 | 153.3869 | AURKA | 135.4749 |
| NCAPG | 152.5039 | KNG1 | 132.7054 |
| OPRD1 | 151.2574 | GAL | 129.4619 |
| DLGAP5 | 144.4411 | CDC45 | 128.8082 |
3.2 Significance of candidate ego-network modules
To evaluate the statistical significance of candidate egonetwork modules, we carried out random permutation test 1000 times to obtain the null distribution of classification accuracy. After random permutation test with 1000 times, we found that a total of 5 modules including module 59, 72, 78, 86, and 90 were significant with FDR < 0.01 in permutation tests (Table 2). The composition of these differential ego-network modules was exhibited in Figure 2.
The 5 differential ego-network modules with false discovery rate (FDR) < 0.01, and the ego genes, as well as classification accuracy in these modules.
| Modules | Ego gene | Classification accuracy | FDR |
|---|---|---|---|
| Module 59 | ADCY2 | 1.00 | 0 |
| Module 72 | BIRC5 | 0.90 | 0 |
| Module 78 | PRKCA | 1.00 | 0 |
| Module 86 | DTL | 1.00 | 0 |
| Module 90 | HMMR | 0.96 | 0 |
Identified differential ego-network modules in glioma and their composition. A, B, C, D, and E stood for ego-network modules 59, 72, 78, 86, and 90, respectively. Yellow nodes demonstrated the ego genes.
3.3 Pathway annotation analysis
When the pathways with less than 5 genes or more than 100 genes were removed, 1135 pathways were reserved for subsequent enrichment analysis. After multiple tests, we obtained the pathways enriched by the genes of differential ego-network modules, as listed in Table 3. From this table, we found that the genes in module 90 were involved in the pathway of resolution of sister chromatid cohesion (FDR = 5.58E-09), the genes in module 86 participated in the pathway of mitotic prometaphase (FDR = 8.64E-08), the genes of module 59 were enriched in the pathway of G alpha (s) signaling events (FDR = 6.11E-06), the genes in module 72 participated in the pathway of SOS-mediated signaling (FDR = 3.65E-04), and the genes of module 78 were involved in the pathway of VEGFA-VEGFR2 pathway (FDR = 1.09E-04).
Pathways enriched by the genes of differential ego-network modules based on FDR < 0.05
| Modules | Pathways | FDR | Ego genes |
|---|---|---|---|
| Module 90 | Resolution of sister chromatid cohesion | 5.58E-09 | HMMR |
| Module 86 | Mitotic prometaphase | 8.64E-08 | DTL |
| Module 59 | G alpha (s) signaling events | 6.11E-06 | ADCY2 |
| Module 72 | SOS-mediated signaling | 3.65E-04 | BIRC5 |
| Module 78 | VEGFA-VEGFR2 pathway | 1.09E-04 | PRKCA |
4 Discussion
Glioma is the most common type of primary brain cancer. In the ongoing effort to characterize this disease, more and more studies have indicated correlations between the pathogenesis of glioma and specific genetic alterations. Of note, these studies have shown considerable variability in the genetic factors related with glioma. This diversity underscores the demand for more valuable and accurate signatures, and putative targets for developing novel therapeutic paradigms [24, 25]. Thus, in the present study, we constructed a DCN including 6325 edges among 2184 nodes for glioma, and based on this DCN, a total of 109 genes were selected as ego genes. Identification of ego-network modules was expanded on the ego genes according to the increased classification accuracy using the EgoNet algorithm. After evaluating by random permutation testing, we obtained 5 differential egonetwork modules with FDR < 0.01 between glioma and control conditions. Significantly, the genes of module 90 and 86 were enriched in the pathway of resolution of sister chromatid cohesion and mitotic prometaphase, respectively.
Cohesin, well conserved across organisms, is a multisubunit complex which is related with chromatin following mitosis and is crucial for holding sister chromatids together after DNA replication [26]. A former study has indicated that RNAi-mediated down-regulation of SMC1A (known as a cohesion subunit) leads to chromatid cohesion defects in human cells [27]. Importantly, another study has reported that chromosome cohesion defects might contribute to aneuploidy in human cancer cells, the most common characteristic of human solid tumors [28]. Frequent alterations in genes involved in sister chromatid cohesion have been observed in bladder cancer [29]. Moreover, a sister chromatid cohesion protein (Pds5A) has been demonstrated to be up-regulated in high grade gliomas [30]. More importantly, in the process of mitosis, mitotic spindle checkpoint ensures the fidelity of sister chromatid segregation through delaying the onset of anaphase until the kinetochores of all the chromosomes have attached to the spindle microtubules [31]. Abnormal function of a kinetochore results in the gains or losses of large portions of chromosomes [32]. In addition, chromosomal instability is a definitive characteristic in tumor progression. In this study, the genes of module 90 and 86 were enriched in the pathway of resolution of sister chromatid cohesion and mitotic prometaphase, respectively. Thus, we infer that module 90 and 86, and their respective pathway of resolution of sister chromatid cohesion and mitotic prometaphase might play important role in the development of glioma.
As reported, HMMR is associated with/demonstrates tumorigenicity in glioblastoma stem cells [33]. Moreover, Ying and colleagues have indicated that HMMR is overexpressed in glioblastoma samples of human [34]. These suggest that HMMR might be an oncogene. Another study has demonstrated that up-regulation of CCNB1 is correlated with poor outcomes in glioma patients [35]. Significantly, BIRC5 exhibited the positive expression in glioblastoma samples [36]. Interestingly, HMMR was the ego gene in the module 90 in our study. Of note, the other genes CCNB1, BIRC5 and ego gene HMMR were clustered in one ego-network module 90. This finding was in line with the local property of disease networks, that is to say, proteins related with the same disease had a tendency to interact with each other [37]. It was noteworthy that BIRC5 was the ego gene in the module 72 in our study. Thus, we infer that BIRC5 might be a pivotal factor in glioma development. However, follow-up experiments are needed to verify specific links between these findings and glioma.
In conclusion, These results showed that the core methodology introduced in this paper, including EgoNet algorithm and the accompanying random permutation test for assessing statistical significance, was capable of identifying ego-network modules of coordinately expressed genes which pointed to key regulators in disease networks and thus provided a more systematic understanding of complex disease progression. The above discussed modules may be regarded as potential disease modules whose activity dictates glioma progression. The mechanisms might imply that ego genes including HMMR and BIRC5 are key factors to distinguish glioma patients. However, our study was conducted using existing data based on bioinformatics methods, yet the findings lacked experimental validations. Therefore, further investigations are needed to discover the alterations of these modules in the understanding of the molecular mechanisms underlying glioma based on the experiments using animal or patient tissues.
Conflict of interest: Authors state no conflict of interest.
References
[1] Wen P.Y., Kesari S., Malignant gliomas in adults, New Engl J. Med., 2008, 359, 492-50710.1056/NEJMra0708126Search in Google Scholar PubMed
[2] Louis D.N., Ohgaki H., Wiestler O.D., Cavenee W.K., Burger P.C., Jouvet A., et al., The 2007 WHO classification of tumours of the central nervous system, Acta Neuropathol., 2007, 114, 97-10910.1007/s00401-007-0243-4Search in Google Scholar PubMed PubMed Central
[3] Yamanaka R., Cell-and peptide-based immunotherapeutic approaches for glioma, Trends Mol. Med., 2008, 14, 228-23510.1016/j.molmed.2008.03.003Search in Google Scholar PubMed
[4] García-Claver A., Lorente M., Mur P., Campos-Martín Y., Mollejo M., Velasco G., et al., Gene expression changes associated with erlotinib response in glioma cell lines, Eur. J. Cancer., 2013, 49, 1641-165310.1016/j.ejca.2013.01.002Search in Google Scholar PubMed
[5] Zhu X.L., Ai Z.H., Wang J., Xu Y.L. Teng Y.C., Weighted gene co-expression network analysis in identification of endometrial cancer prognosis markers, Asian Pac. J. Cancer P., 2012, 13, 4607-461110.7314/APJCP.2012.13.9.4607Search in Google Scholar
[6] Venet D., Dumont J.E., Detours V., Most random gene expression signatures are significantly associated with breast cancer outcome, PLoS Comput. Biol., 2011, 7, e100224010.1371/journal.pcbi.1002240Search in Google Scholar PubMed PubMed Central
[7] He D., Liu Z.-P., Honda M., Kaneko S. Chen L., Coexpression network analysis in chronic hepatitis B and C hepatic lesions reveals distinct patterns of disease progression to hepatocellular carcinoma, J. Mol. Cell Biol., 2012, 4, 140-15210.1093/jmcb/mjs011Search in Google Scholar PubMed
[8] Liu Z.P., Wang Y., Zhang X.S. and Chen L.N., Network-based analysis of complex diseases, IET Syst. Biol., 2012, 6, 22-3310.1049/iet-syb.2010.0052Search in Google Scholar PubMed
[9] Nibbe R.K., Chowdhury S.A., Koyutürk M., Ewing R. Chance M.R., Protein–protein interaction networks and subnetworks in the biology of disease, WIRES Syst. Biol. Med., 2011, 3, 357-36710.1002/wsbm.121Search in Google Scholar PubMed
[10] Wu Y., Jing R., Jiang L., Jiang Y., Kuang Q., Ye L., et al., Combination use of protein–protein interaction network topological features improves the predictive scores of deleterious non-synonymous single-nucleotide polymorphisms, Amino acids., 2014, 46, 2025-203510.1007/s00726-014-1760-9Search in Google Scholar PubMed
[11] Liu Z.P., Wang Y., Zhang X.S., Chen L.N., Network-based analysis of complex diseases, Systems Biology, IET., 2012, 6, 22-3310.1049/iet-syb.2010.0052Search in Google Scholar PubMed
[12] Chen L., Xuan J., Riggins R.B., Wang Y., Clarke R., Identifying protein interaction subnetworks by a bagging Markov random field-based method, Nucleic Acids Res., 2013, 41, e42-e4210.1093/nar/gks951Search in Google Scholar PubMed PubMed Central
[13] Ciriello G., Cerami E., Sander C., Schultz N., Mutual exclusivity analysis identifies oncogenic network modules, Genome Res., 2012, 22, 398-40610.1101/gr.125567.111Search in Google Scholar PubMed PubMed Central
[14] Su J., Yoon B.J., Dougherty E.R., Identification of diagnostic subnetwork markers for cancer in human protein-protein interaction network, BMC Bioinformatics., 2010, 11, 110.1186/1471-2105-11-S6-S8Search in Google Scholar PubMed PubMed Central
[15] Yang R., Bai Y., Qin Z., Yu T., EgoNet: identification of human disease ego-network modules, BMC genomics., 2014, 15, 31410.1186/1471-2164-15-314Search in Google Scholar PubMed PubMed Central
[16] Borgatti S.P., Mehra A., Brass D.J., Labianca G., Network analysis in the social sciences, Science., 2009, 323, 892-89510.1126/science.1165821Search in Google Scholar PubMed
[17] Gautier L., Cope L., Bolstad B.M., Irizarry R.A., affy—analysis of Affymetrix GeneChip data at the probe level, Bioinformatics., 2004, 20, 307-31510.1093/bioinformatics/btg405Search in Google Scholar PubMed
[18] Tusher V.G., Tibshirani R., Chu G., Significance analysis of microarrays applied to the ionizing radiation response, P Natl A Sci., 2001, 98, 5116-512110.1073/pnas.091062498Search in Google Scholar PubMed PubMed Central
[19] Golembieski W.A., Rempel S.A., cDNA array analysis of SPARC-modulated changes in glioma gene expression, J. Neuro-oncol., 2002, 60, 213-22610.1023/A:1021167211131Search in Google Scholar
[20] Jensen L.J., Kuhn M., Stark M., Chaffron S., Creevey C., Muller J., et al., STRING 8—a global view on proteins and their functional interactions in 630 organisms, Nucleic Acids Res., 2009, 37, D412-D41610.1093/nar/gkn760Search in Google Scholar PubMed PubMed Central
[21] Goodman L.A., Snowball sampling, The annals of mathematical statistics., 1961, 148-17010.1214/aoms/1177705148Search in Google Scholar
[22] Benjamini Y., Hochberg Y., Controlling the false discovery rate: a practical and powerful approach to multiple testing, J. R. Stat. Soc., 1995, 289-30010.1111/j.2517-6161.1995.tb02031.xSearch in Google Scholar
[23] Ahn T., Lee E., Huh N., Park T., Personalized identification of altered pathways in cancer using accumulated normal tissue data, Bioinformatics., 2014, 30, i422-i42910.1093/bioinformatics/btu449Search in Google Scholar PubMed PubMed Central
[24] Louis D.N., Molecular pathology of malignant gliomas, Annu. Rev. Pathol. Mech. Dis., 2006, 1, 97-11710.1146/annurev.pathol.1.110304.100043Search in Google Scholar PubMed
[25] Sathornsumetee S., Rich J.N., Designer therapies for glioblastoma multiforme, Ann NY Acad. Sci., 2008, 1142, 108-13210.1196/annals.1444.009Search in Google Scholar PubMed
[26] Dorsett D., Ström L., The ancient and evolving roles of cohesin in gene expression and DNA repair, Curr Biol., 2012, 22, R240-R25010.1016/j.cub.2012.02.046Search in Google Scholar PubMed PubMed Central
[27] Barber T.D., McManus K., Yuen K.W., Reis M., Parmigiani G., Shen D., et al., Chromatid cohesion defects may underlie chromosome instability in human colorectal cancers, P. Natl. A. Sci., 2008, 105, 3443-344810.1073/pnas.0712384105Search in Google Scholar PubMed PubMed Central
[28] Kops G.J., Weaver B.A., Cleveland D.W., On the road to cancer: aneuploidy and the mitotic checkpoint, Nat. Rev. Cancer., 2005, 5, 773-78510.1038/nrc1714Search in Google Scholar PubMed
[29] Guo G., Sun X., Chen C., Wu S., Huang P., Li Z., et al., Whole-genome and whole-exome sequencing of bladder cancer identifies frequent alterations in genes involved in sister chromatid cohesion and segregation, Nat. Genet., 2013, 45, 1459-146310.1038/ng.2798Search in Google Scholar PubMed PubMed Central
[30] Hagemann C., Weigelin B., Schommer S., Schulze M., Al-Jomah N., Anacker J., et al., The cohesin-interacting protein, precocious dissociation of sisters 5A/sister chromatid cohesion protein 112, is up-regulated in human astrocytic tumors, Cancer., 2011, 21, 2210.3892/ijmm.2010.551Search in Google Scholar PubMed
[31] Bharadwaj R., Yu H., The spindle checkpoint, aneuploidy, and cancer, Oncogene., 2004, 23, 2016-202710.1038/sj.onc.1207374Search in Google Scholar PubMed
[32] Lengauer C., Kinzler K.W., Vogelstein B., Genetic instabilities in human cancers, Nature., 1998, 396, 643-64910.1038/25292Search in Google Scholar PubMed
[33] Tilghman J., Wu H., Sang Y., Shi X., Guerrero-Cazares H., Quinones-Hinojosa A., et al., HMMR maintains the stemness and tumorigenicity of glioblastoma stem-like cells, Cancer Res., 2014, 74, 3168-317910.1158/0008-5472.CAN-13-2103Search in Google Scholar PubMed PubMed Central
[34] Ying M., Tilghman J., Laterra J., Hyaluronan-mediated motility receptor as a novel target for inhibiting glioblastoma stem cells, Cancer Res., 2012, 72, 3476-347610.1158/1538-7445.AM2012-3476Search in Google Scholar
[35] de Haas T., Hasselt N., Troost D., Caron H., Popovic M., Zadravec-Zaletel L., et al., Molecular risk stratification of medulloblastoma patients based on immunohistochemical analysis of MYC, LDHB, and CCNB1 expression, Clin. Cancer Res., 2008, 14, 4154-416010.1158/1078-0432.CCR-07-4159Search in Google Scholar PubMed
[36] Jung T.Y., Choi Y.D., Kim Y.H., Lee J.J., Kim H.S., Kim J.S., et al., Immunological characterization of glioblastoma cells for immunotherapy, Anticancer Res., 2013, 33, 2525-2533Search in Google Scholar
[37] Barabási A.L., Gulbahce N., Loscalzo J., Network medicine: a network-based approach to human disease, Nat. Rev. Genet., 2011, 12, 56-6810.1038/nrg2918Search in Google Scholar PubMed PubMed Central
© 2017 Li He et al.
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
