SWIM tool application to expression data of glioblastoma stem-like cell lines, corresponding primary tumors and conventional glioma cell lines
It is well-known that glioblastoma contains self-renewing, stem-like subpopulation with the ability to sustain tumor growth. These cells – called cancer stem-like cells – share certain phenotypic characteristics with untransformed stem cells and are resistant to many conventional cancer therapies, which might explain the limitations in curing human malignancies. Thus, the identification of genes controlling the differentiation of these stem-like cells is becoming a successful therapeutic strategy, owing to the promise of novel targets for treating malignancies.
Recently, we developed SWIM, a software able to unveil a small pool of genes – called switch genes – critically associated with drastic changes in cell phenotype. Here, we applied SWIM to the expression profiling of glioblastoma stem-like cells and conventional glioma cell lines, in order to identify switch genes related to stem-like phenotype.
SWIM identifies 171 switch genes that are all down-regulated in glioblastoma stem-like cells. This list encompasses genes like CAV1, COL5A1, COL6A3, FLNB, HMMR, ITGA3, ITGA5, MET, SDC1, THBS1, and VEGFC, involved in “ECM-receptor interaction“ and “focal adhesion” pathways. The inhibition of switch genes highly correlates with the activation of genes related to neural development and differentiation, such as the 4-core OLIG2, POU3F2, SALL2, SOX2, whose induction has been shown to be sufficient to reprogram differentiated glioblastoma into stem-like cells. Among switch genes, the transcription factor FOSL1 appears as the brightest star since: it is down-regulated in stem-like cells; it highly negatively correlates with the 4-core genes that are all up-regulated in stem-like cells; the promoter regions of the 4-core genes harbor a consensus binding motif for FOSL1.
We suggest that the inhibition of switch genes in stem-like cells could induce the deregulation of cell communication pathways, contributing to neoplastic progression and tumor invasiveness. Conversely, their activation could restore the physiological equilibrium between cell adhesion and migration, hampering the progression of cancer. Moreover, we posit FOSL1 as promising candidate to orchestrate the differentiation of cancer stem-like cells by repressing the 4-core genes’ expression, which severely halts cancer growth and might affect the therapeutic outcome. We suggest FOSL1 as novel putative therapeutic and prognostic biomarker, worthy of further investigation.
KeywordsBioinformatics Cancer stem cell Graph theory Glioblastoma
Average pearson correlation coefficient
Differentially expressed genes
False discovery rate
Gene expression omnibus
Glioblastoma full stem-like phenotype
Graphical user interphase
Inter quartile range
Sum of the squared error
The cancer genome atlas
World health organization
Glioblastoma multiforme (GBM) is the most frequently diagnosed brain tumor in adults [1, 2] according to the World Health Organization (WHO). Recently, the unique professional research organization CBTRUS provided a comprehensive summary of the current descriptive epidemiology of primary brain and central nervous system tumors in the United States population for diagnosis years 2008-2012 . From this prospective study emerges that glioblastoma accounts for 15.1% of all primary brain tumors and 46.1% of primary malignant brain tumors; it is more common in older adults especially in males (is about 1.6 times higher in males as compared to females), and is less common in children; it has the highest incidence among all malignant tumors, with 11890 cases predicted in 2015 and 12120 in 2016.
Current scientific research and clinical trials have not led to a definitive cure for GBM but have contributed to both an improved understanding of the disease progression, as well as small improvements in patient outcomes to treatment. In particular, several studies identified a small percentage of the total GBM cell population that evolves along the course of the disease, forming highly heterogeneous subpopulations within the tumor mass. These cells possess radio/chemo-resistant properties and may have a role in driving tumor initiation, resistance to treatment, tumor progression, and relapse [10, 11, 12, 13, 14, 15, 16]. Due to their ability of self-renewing, proliferating, and differentiating into multiple lineages, this subpopulation of cells - known as cancer stem-like cells  - is held responsible for carcinogenesis not only in brain cancer [10, 13], but even in other tumors such as breast, colon, prostate, pancreatic, melanoma cancers [10, 13, 18, 19, 20, 21, 22]. The failure in removing these GBM cancer stem-like cells is one of the main reason behind the ineffectiveness of traditional therapies in treating glioblastoma . Therefore, focusing on the characteristics of GBM stem-like cells and on necessary conditions for specific cell differentiation is a promising strategy to propose new therapeutic targets in order to improve GBM treatment efficacy and overcome drug resistance.
Here, we applied SWItchMiner SWIM [23, 24] – a software that we recently developed to unveil a small pool of genes (called switch genes) critically associated with drastic changes in cell phenotype – to the expression data obtained by Affymetrix HG-U133 Plus 2.0 microrarrays of glioblastoma stem-like (GS) cell lines, corresponding primary tumors, and conventional glioma cell lines , publicly available on the Gene Expression Omnibus (GEO) repository . Our aim was to identify switch genes in the transition from stem-like to differentiated GBM cells.
Schulte et al.
Verhaak et al.
The second GBM dataset analyzed for the present study is available as supplementary material of a recent study. Data include microarray expression profiles of 173 core TCGA samples unified and scaled from three gene expression platforms (Affymetrix HuEx array, Affymetrix U133A array, and Agilent 244K array). The authors of this study described a robust gene expression-based molecular classification of GBM into Proneural, Neural, Classical, and Mesenchymal subtypes and integrate multidimensional genomic data to establish patterns of somatic mutations and DNA copy number. Aberrations and gene expression of EGFR, NF1, PDGFRA/IDH1, and neuron markers (e.g. NEFL, GABRA1, SYT1, SLC12A5) each define the Classical, Mesenchymal, Proneural and Neural subtypes, respectively.
The third GBM dataset analyzed for the present study was downloaded from TCGA Data Portal Release 10.0 (December 2017) [8, 9]. It represents GBM normalized expression data of 161 GBM unique patients from high-throughput RNA sequencing, created by using FPKM procedure to perform the normalization (i.e. HTSeq-FPKM data). For these patients also clinical data were downloaded from TCGA in order to perform the Kaplan-Meier survival analysis.
SWItchMiner (SWIM)  is a software with a user-friendly Graphical User Interphase (GUI) developed in MATLAB and downloadable from the supplementary materials of . SWIM implements an integrated network analysis able to extract from genome-wide expression data key players (i.e. switch genes) marking the shift from one condition to another in a complex biological network. SWIM algorithm encompasses a series of well-defined steps described in the following .
Step 1: Pre-processing phase
Denoting by A and B the two conditions between which searching for switch genes and by S the total number of samples (S = samples in the condition A + samples in the condition B), this step requires the selection of two specific thresholds for removing genes whose expression across the S samples is mostly zero or change very little. The first threshold regards the maximum number of samples out of S allowed to be equal to zero. The second threshold concerns the minimum variation - measured by the Inter Quartile Range (IQR) percentile - allowed for each gene across the S samples.
Step 2: Filtering phase phase
This step requires the selection of two specific thresholds for removing genes whose expression between the two given conditions (A and B) does not change enough or does it without statistical significance. Considering the logarithm of the ratio between the average expression of samples in condition A and the average expression of samples in condition B (log fold-change), the first threshold allows to remove the genes falling behind, in absolute value, a fixed cutoff on the log fold-change. The second threshold concerns the smallest probability (p-value) for which the data allow to reject the null hypothesis (i.e. the means of the two distributions – normal and cancer - are identical) of the Student’s t-test. Actually, since this statistical test will be repeated multiple times (as many as the genes under testing), the obtained p-values must be adjusted. To correct multiple tests, SWIM makes use of False Discovery Rate (FDR) method  and thus the threshold refers to the FDR values. At end of this phase, the differentially expressed genes between conditions A and B have been identified.
Step 3: Building the correlation network
This step requires the selection of a threshold for building the correlation network where two nodes are connected if the absolute value of the Pearson correlation between their expression profiles exceeds a given cutoff. This threshold should reflect a right balance between the number of edges and the number of connected components of the network: the number of edges should be as small as possible in order to have a manageable network (pointing towards a higher threshold) and the number of connected components should be as small as possible in order to preserve the integrity of the network (pointing towards a smaller threshold).
Step 4: Finding communities in the network
Step 5: Building the heat cartography map
- 1non local hub for zg<2.5
Kπ=0 Ultra-peripheral nodes (role R1)
Kπ≤0.625 Peripheral nodes (role R2)
0.62<Kπ≤0.8 Non-hub connectors (role R3)
Kπ>0.8 Non-hub kinless nodes (role R4)
- 2local hub for zg≥2.5
Kπ=0.3 Provincial hubs (role R5)
Kπ≤0.75 Connector hubs (role R6)
Kπ>0.75 Kinless hubs (role R7)
Then, SWIM colors nodes in the cartography according to the Average Pearson Correlation Coefficient (APCC) between the expression profiles of each node and its nearest neighbors . This representation of the network is defined as “heat cartography map”. By computing the APCC of expression over all interaction partners of each hub in protein-protein interaction (PPI) networks in yeast, the authors in  concluded that hubs fall into two distinct categories: date hubs that display low co-expression with their partners (low APCC) and party hubs that have high co-expression (high APCC). In the gene expression networks, the distribution of APCCs appears to be trimodal [23, 24] where, similarly to PPI networks, two peaks represent low (date hubs) and high (party hubs) positive APCC values, but with the addition of a new third peak which is characteristic of gene expression networks and represents negative APCC values. Nodes populating this peak are called “fight-club hubs”.
Step 6: Identification of switch genes
being not an hub in their own cluster (zg<2.5);
having many links outside their own cluster (Kpi>0.8);
having a negative average weight of their incident links (APCC < 0).
At the end of step 6, SWIM gives the opportunity to perform further analyses regarding the evaluation of network robustness, which is the resilience to errors, by studying the effect on the network connectivity of removing nodes by decreasing degree. In particular, SWIM evaluates the effect on the average shortest path - where the shortest path between two nodes is the minimum number of edges connecting them and the average shortest path is the mean of the shortest paths for all possible pairs of nodes in the network - of removing randomly chosen nodes, switch genes, fight-club hubs, date and party hubs. Since scale-free networks have few hubs and many non-hub nodes, they are amazingly resistant to a random removal of nodes, while the removal of hubs causes an effect known as “vulnerability to attack” to allude to the fact that the integrity of the network is destroyed.
Functional and motif enrichment analysis
The associations between selected genes and functional annotation terms such as Gene Ontology (GO) terms  and KEGG pathways  were analyzed by using FIDEA web tool . Binding motif enrichment analysis in promoter regions (identified as genomic regions spanning from -450 to +50 nucleotides with respect to transcription start sites) was performed by Pscan , which employs the JASPAR 2018 motif collection . A p-value < 0.05, after adjustment for multiple testing performed with the Benjamini-Hochberg method , was set as threshold to identify functional annotations and regulatory motifs significantly enriched amongst the selected gene lists.
microRNA target enrichment analysis
Network visualization and analysis
The free software package Cytoscape was used for visualizing gene correlation networks . To find modules (i.e. locally dense regions) in the gene correlation network, we made use of the Cytoscape plugin MCODE , which weights nodes by a local neighborhood density measure and graphically displays ranked extracted modules.
In order to evaluate the clinical relevance of switch genes identified by SWIM, we performed Kaplan-Meier analysis  by using clinical and RNA-seq expression data provided by TCGA Data Portal Release 10.0 (December 2017) [8, 9], relating to 161 unique GBM patients and GBM subtype-specific patients. The patients were split into two groups (called low-expression and high-expression) according to the expression level of each switch gene. In particular, low- and high-expression groups referred to patients with expression levels lower than or greater than the 50th percentile, respectively. For each patient cohort, the cumulative survival rates were computed according to the Kaplan-Meier method . A log-rank test was performed to evaluate the p-value: the lower the p-value, the better the separation between the prognoses of the two groups. The resulting p-values were adjusted for multiple testing by using the Benjamini-Hochberg (FDR) procedure .
Integrated network analysis of genes involved in the transition from glioblastoma stem-like to conventional cell lines reveals fight-club hubs
SWIM next searched for specific topological properties of the correlation network using the date/party/fight-club hub classification system, which we previously defined [23, 24], based on the Average Pearson Correlation Coefficients (APCCs) between the expression profiles of each hub and its nearest neighbors. The extent to which hubs are co-expressed with their interaction partners leads to three classifications with characteristic topological properties: date hubs (low positive APCC), party hubs (high positive APCC), and fight-club hubs (negative APCC). Date hubs have a coordinating role within the network, whereas party hubs act as local hubs . Likewise date hubs, fight-club hubs are supposed to connect different biological processes, thus acting as global hubs, but differently from them they display an opposite transcriptional pattern with respect to their interaction partners: if they are induced, their interaction partners are repressed, and viceversa.
Heat cartography in glioblastoma reveals switch genes as network bottlenecks
SWIM next searched for the communities within the glioblastoma correlation network using k-means clustering algorithm (see step 4 of SWIM software subsection of Methods), which led to the identification of three clusters or modules (Additional file 4). The intramodule and intermodule connections were exploited by SWIM in order to assign topological roles to each node  based on the computation of two parameters for each node: the clusterphobic coefficient Kπ, which measures the “fear” of each node of being confined in a cluster in analogy with the claustrophobic disorder, and the global within-module degree zg, which measures how “well-connected” each node is to other nodes in its own community. In particular, high zg values correspond to nodes that are hubs within their module (local hubs), while high values of Kπ identify nodes that interact mainly outside their community (see step 5 of SWIM software subsection of Methods). The values of these two parameters allow to define the heat cartography map for the glioblastoma dataset, where party, date, and fight-club hubs were easily identified by red, orange, and blue coloring, respectively (Fig. 5b). Fight-club hubs, acting as negative regulators, mainly fall in the so-called R4 region of the heat cartography map that is characterized by high values of the clusterphobic coefficient and by a strong inclination of nodes to interact mostly outside their own community. This subset of fight-club hubs lying in the region R4 was called switch genes (see step 6 of SWIM software subsection of Methods).
Parameters’ summary used by SWIM for Schulte et al. GBM dataset 
Correlation network nodes
Correlation network edges
Characterization of switch genes
Next, we investigated possible co-regulation of the four core TFs by using Pscan , that evaluates enrichment of known binding motifs in promoter regions, employing the JASPAR 2018 motif collection . Significant enrichment was found for FOSL1 that, thus, resulted as a putative transcription factor binding to four core TFs regulatory elements (Fig. 7b).
Recently, the expression of FOSL1 has been linked to focal adhesion closing thus the circle with the the results of the functional enrichment analysis of the switch genes that reported “ECM-receptor interaction” and “focal adhesion” as the most over-represented pathways. It has been suggested that, in a mouse model of embryonic development in vitro, FOSL1 functions as a modulator of the level of key molecules on endothelial cell surface. It can function as either an activator or a repressor, depending on the gene-context, controlling in this way the delicate equilibrium between adhesion and migration in sprouting angiogenesis .
Taken together these findings thrust FOSL1 into the spotlights as the most promising candidate among switch genes as novel therapeutic target for treating human glioma.
microRNAs regulating switch genes
In order to elucidate the cascade of events underlying the maintenance of the glioblastoma stem-like cells identity, we surveyed regulatory activity of miRNAs on switch genes as computationally predicted by TargetScan  and experimentally validated by miRTarBase .
We found that all these top-ranked miRNAs target those switch genes that are involved in “ECM-receptor interaction” and “focal adhesion” pathways - such as COL5A1, COL6A3, FLNB, ITGA5, MET, THBS1, SDC1, VEGFC - suggesting a further layer of regulation given by miRNAs that could inhibit the “ECM-receptor interaction” and “focal adhesion” pathways by directly targeting switch genes involved in them, and thus promoting cancer invasion and migration.
Similar to the above, we performed the same enrichment analysis for the experimentally validated miRNA-target interaction and once again the miR-26 family and the miR-144-3p appear as the top-ranked miRNAs (Fig. 10b).
Topological relevance of fight-club hubs and switch genes in glioblastoma correlation networks
Evaluating the contribute of switch genes to the robustness of the network, we found the same behavior observed on the deletion of fight-club hubs (Fig. 11b) and the same drastic effect upon removal the first 40% of switch genes, as expected because 98% of them are switch genes. This crucial subset of switch genes encompasses FLNB, ITGA3, MET, THBS1, VEGFC, thus resulting enriched in “ECM-receptor interaction” and “focal adhesion” pathways, and also FOSL1, whose function is related to these pathways and which we acclaimed as the most promising GBM switch gene.
Nowadays, the GBM research scene is dominated by trying to discover novel therapeutic and prognostic markers promoting the differentiation of cancer stem-like cells, which severely halts cancer growth and might affect the therapeutic outcome. Although the adjusted p-values for multiple testing didn’t reveal any statistically significant association (p-value < 0.05) of switch genes with patient survival, the transcription factor FOSL1 fulfills very interesting features that make it eligible as new potential therapeutic target. In particular: it was found to act as repressor transcription factor ; it resulted down-regulated in stem-like cells; it resulted highly negatively correlated with the 4-core TFs that were resulted all up-regulated in stem-like cells; the promoter regions of the 4-core TFs were found to harbor a consensus binding motif for FOSL1. Taken together these considerations prompt us to bet on FOSL1, which can promote the differentiation process of GBM stem-like cells by repressing the 4-core TFs. This should allow for anticipation of care as well as the reduction of the social impact of diseases and the restraint of health costs.
Although our study can be considered as a starting point, and further functional and clinical investigations are needed, the switch gene signatures and their nearest neighbor genes can improve our knowledge of the cellular events that are crucial for carcinogenesis and they also reveal many potential prognostic and novel therapeutic targets that have so far not been linked to glioblastoma. Thus, using SWIM could provide important clues that will stimulate research activities into the causes of this terrible disease thus supporting the planning of healthcare services such as clinical trials and disease prevention.
Publication costs for this manuscript were sponsored by SysBioNet, Italian Roadmap Research Infrastructures 2012.
Availability of data and materials
The results shown in this paper are in part based upon data available at GEO under accession number GSE23806 published on Feb 12, 2011 by .
About this supplement
This article has been published as part of BMC Bioinformatics Volume 19 Supplement 15, 2018: Proceedings of the 12th International BBCC conference. The full contents of the supplement are available online at https://bmcbioinformatics.biomedcentral.com/articles/supplements/volume-19-supplement-15.
PP conceived and designed the research. PP developed the software. GF performed computational data analysis and prepared figures. FC performed the results interpretation of the computational analysis. All authors wrote the manuscript. All authors have read and approved the final manuscript.
Ethics approval and consent to participate
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 3.Ostrom QT, Gittleman H, Fulop J, Liu M, Blanda R, Kromer C, Wolinsky Y, Kruchko C, Barnholtz-Sloan JS. Cbtrus statistical report: Primary brain and central nervous system tumors diagnosed in the united states in 2008-2012. Neuro-Oncol. 2015; 17(suppl 4):1–62.Google Scholar
- 4.Grossman SA, Ye X, Piantadosi S, Desideri S, Nabors LB, Rosenfeld M, Fisher J, Consortium NC, et al. Survival of patients with newly diagnosed glioblastoma treated with radiation and temozolomide in research studies in the united states. Clin Cancer Res. 2010; 16(8):2443–9.PubMedPubMedCentralGoogle Scholar
- 5.Mizoe J-E, Tsujii H, Hasegawa A, Yanagi T, Takagi R, Kamada T, Tsuji H, Takakura K, of the Central Nervous System Tumor Working Group OC, et al. Phase i/ii clinical trial of carbon ion radiotherapy for malignant gliomas: combined x-ray radiotherapy, chemotherapy, and carbon ion radiotherapy. Int J Radiat Oncol* Biol* Phys. 2007; 69(2):390–6.Google Scholar
- 9.McLendon R, Friedman A, Bigner D, Van Meir EG, Brat DJ, Mastrogianakis GM, Olson JJ, Mikkelsen T, Lehman N, Aldape K, et al. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature. 2008; 455(7216):1061–8.Google Scholar
- 17.Guo W, Lasky JL, Wu H. Cancer stem cells. Pediatr Res. 2006; 59:59–64.Google Scholar
- 23.Paci P, Colombo T, Fiscon G, Gurtner A, Pavesi G, Farina L. Swim: a computational tool to unveiling crucial nodes in complex biological networks. Sci Rep. 2016; 7:44797.Google Scholar
- 24.Palumbo MC, Zenoni S, Fasoli M, Massonnet M, Farina L, Castiglione F, Pezzotti M, Paci P. Integrated network analysis identifies fight-club nodes as a class of hubs encompassing key putative switch genes that induce major transcriptome reprogramming during grapevine development. Plant Cell. 2014; 26(12):4617–35.PubMedPubMedCentralGoogle Scholar
- 25.Schulte A, Günther HS, Phillips HS, Kemming D, Martens T, Kharbanda S, Soriano RH, Modrusan Z, Zapf S, Westphal M, et al. A distinct subset of glioma cell lines with stem cell-like properties reflects the transcriptional phenotype of glioblastomas and overexpresses cxcr4 as therapeutic target. Glia. 2011; 59(4):590–602.PubMedGoogle Scholar
- 26.Barrett T, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M, Marshall KA, Phillippy KH, Sherman PM, Holko M, et al. Ncbi geo: archive for functional genomics data sets—update. Nucleic Acids Res. 2013; 41(D1):991–5.Google Scholar
- 27.Verhaak RG, Hoadley KA, Purdom E, Wang V, Qi Y, Wilkerson MD, Miller CR, Ding L, Golub T, Mesirov JP, et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in pdgfra, idh1, egfr, and nf1. Cancer Cell. 2010; 17(1):98–110.PubMedPubMedCentralGoogle Scholar
- 28.Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Statist Soc B. 1995; 57(1):289–300.Google Scholar
- 29.Hartigan JA, Wong MA. Algorithm as 136: A k-means clustering algorithm. J R Stat Soc Ser C (Appl Stat). 1979; 28(1):100–8.Google Scholar
- 30.Meilă M. The uniqueness of a good optimum for k-means. In: Proceedings of the 23rd International Conference on Machine Learning. New York: ACM: 2006. p. 10121–0701.Google Scholar
- 31.Lisboa PJ, Etchells TA, Jarman IH, Chambers SJ. Finding reproducible cluster partitions for the k-means algorithm. BMC Bioinformatics. 2013; 14(1):1.Google Scholar
- 35.Kanehisa M, Sato Y, Kawashima M, Furumichi M, Tanabe M. Kegg as a reference resource for gene and protein annotation. Nucleic Acids Res. 2016; 44(D1):1070.Google Scholar
- 36.D’Andrea D, Grassi L, Mazzapioda M, Tramontano A. Fidea: a server for the functional interpretation of differential expression analysis. Nucleic Acids Res. 2013; 41(W1):84–88.Google Scholar
- 37.Zambelli F, Pesole G, Pavesi G. Pscan: finding over-represented transcription factor binding site motifs in sequences from co-regulated or co-expressed genes. Nucleic Acids Res. 2009; 37(suppl 2):247–52.Google Scholar
- 38.Khan A, Fornes O, Stigliani A, Gheorghe M, Castro-Mondragon JA, van der Lee R, Bessy A, Chèneby J, Kulkarni SR, Tan G, Baranasic D, Arenillas DJ, Sandelin A, Vandepoele K, Lenhard B, Ballester B, Wasserman WW, Parcy F, Mathelier A. Jaspar 2018: update of the open-access database of transcription factor binding profiles and its web framework. Nucleic Acids Res. 2018; 46(D1):260–6. https://doi.org/10.1093/nar/gkx1126.Google Scholar
- 40.Chou C-H, Chang N-W, Shrestha S, Hsu S-D, Lin Y-L, Lee W-H, Yang C-D, Hong H-C, Wei T-Y, Tu S-J, et al. mirtarbase 2016: updates to the experimentally validated mirna-target interactions database. Nucleic Acids Res. 2015; 44(D1):239–47.Google Scholar
- 47.Veeravalli KK, Rao JS. Mmp-9 and upar regulated glioma cell migration. Cell Adhes Migr. 2012; 6(6):509–12.Google Scholar
- 48.Zhang L, Liu H, Mu X, Cui J, Peng Z. Dysregulation of fra1 expression by wnt/ β-catenin signalling promotes glioma aggressiveness through epithelial–mesenchymal transition. Biosci Rep. 2017; 37(2):20160643.Google Scholar
- 49.Galvagni F, Orlandini M, Oliviero S. Role of the ap-1 transcription factor fosl1 in endothelial cells adhesion and migration. Cell Adhes Migr. 2013; 7(5):408–11.Google Scholar
- 52.Jiang Q, Liu Y, Zhang S, Li N, Sun G. Mir-26b suppresses cell proliferation and invasion by targeting cyclooxygenase 2 in human glioblastoma. Oncotarget. 2016. https://doi.org/10.18632/oncotarget.12706.
- 57.Floyd DH, Zhang Y, Dey BK, Kefas B, Breit H, Marks K, Dutta A, Herold-Mende C, Synowitz M, Glass R, Abounader R, Purow BW. Novel anti-apoptotic micrornas 582-5p and 363 promote human glioblastoma stem cell survival via direct inhibition of caspase 3, caspase 9, and bim. PLoS ONE. 2014; 9(5):96239. https://doi.org/10.1371/journal.pone.0096239.Google Scholar
- 60.Kinsella RJ, Kähäri A, Haider S, Zamora J, Proctor G, Spudich G, Almeida-King J, Staines D, Derwent P, Kerhornou A, et al. Ensembl biomarts: a hub for data retrieval across taxonomic space. Database. 2011; 2011:030.Google Scholar
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.