Mesenchymal glioblastoma constitutes a major ceRNA signature in the TGF-β pathway

Abstract

Rationale: Competitive endogenous RNA (ceRNA) networks play important roles in posttranscriptional regulation. Their dysregulation is common in cancer. However, ceRNA signatures have been poorly examined in the invasive and aggressive phenotypes of mesenchymal glioblastoma (GBM). This study aims to characterize mesenchymal glioblastoma at the mRNA-miRNA level and identify the mRNAs in ceRNA networks (micNET) markers and their mechanisms in tumorigenesis. Methods: The mRNAs in ceRNA networks (micNETs) of glioblastoma were investigated by constructing a GBM ceRNA network followed by integration with a STRING protein interaction network. The prognostic micNET markers of mesenchymal GBM were identified and validated across multiple datasets. ceRNA interactions were identified between micNETs and miR181 family members. LY2109761, an inhibitor of TGFBR2, demonstrated tumor-suppressive effects on both primary cultured cells and a patient-derived xenograft intracranial model. Results: We characterized mesenchymal glioblastoma at the mRNA-miRNA level and reported a ceRNA network that could separate the mesenchymal subtype from other subtypes. Six genes (TGFBR2, RUNX1, PPARG, ACSL1, GIT2 and RAP1B) that interacted with each other in both a ceRNA-related manner and in terms of their protein functions were identified as markers of the mesenchymal subtype. The coding sequence (CDS) and 3'-untranslated region (UTR) of TGFBR2 upregulated the expression of these genes, whereas TGFBR2 inhibition by siRNA or miR-181a/d suppressed their expression levels. Furthermore, mesenchymal subtype-related genes and the invasion phenotype could be reversed by suppressing the six mesenchymal marker genes. Conclusions: This study suggests that the micNETs may have translational significance in the diagnosis of mesenchymal GBM and may be novel therapeutic targets.

Keywords: TGFBR2; ceRNA network; mesenchymal subtype; micNETs.

Figure 1

Global landscape of mRNA-miRNA interactions in glioblastoma. (A) Genes that interacted with each other in a ceRNA-dependent manner are presented in a cluster arranged by proneural, neural, classical and mesenchymal subtypes. The ceRNA network included 1525 mRNAs and 188 miRNAs. Several genetic alterations are listed in the middle of the heatmap. NF1 mutations were primarily enriched in the mesenchymal subtype. (B) The ceRNA genes formed a network in which the highly expressed micNETs in the mesenchymal subtype are labeled blue, and genes with low expression are labeled yellow. (C) Functions of micNETs in the blue ceRNA hub or yellow ceRNA hub were profiled by GO and KEGG pathway analyses. (D) The circus plot shows the genomic location of micNETs in the ceRNA network.

Figure 2

Identification of six micNETs as a coregulated hub of mesenchymal glioblastomas. (A) The micNETs with more than 100, 200 or 300 connections are highlighted in the ceRNA network. (B) STRING analysis of micNETs with greater than 300 connections identified six genes (TGFBR2, RUNX1, PPARG, GIT2, ACSL1 and RAP1B) that were mutually connected with each other. (C) The heatmap shows the correlations of the six micNETs: TGFBR2, RUNX1, PPARG, ACSL1, GIT2 and RAP1B.

Figure 3

mRNAs-miRNAs form the hub in the ceRNA network. (A) Integrated circus profile of the ceRNA hub in chromosomes. The inner ring represents 27 core micNETs, whereas the outside ring represents their targeting miRNAs. (B) Six core micNETs and their bridge miRNAs are displayed in a Cytoscape network. (C) Six core micNETs were highly expressed, whereas their targeting miRNAs were lowly expressed in the mesenchymal subtype, as presented in the heatmap. (D-G) Relative mRNAs levels from three independent experiments are shown. The CDS and 3'-UTR of TGFBR2 were cloned into pcDNA3.1 vectors. U87 and LN229 cells were transfected with the indicated plasmids or miRNA mimics for 48 h, and mRNA was extracted for micNET analysis. (H) The scheme of RIP analysis. (I) After overexpression of miR-181a and miR-181d in U87 cells, AGO2 proteins were pulled down by an antibody. The expression levels of TGFBR2 and RUNX1 were measured by real-time PCR assays.

Figure 4

Expression levels of six micNETs predicted the mesenchymal subtype. (A-B) ssGSEA enrichment analysis was employed to evaluate the expression pattern of six micNETs and six TFs in the proneural, neural, classical and mesenchymal subtypes in TCGA Agilent and HiSeq databases. (C-D) ROC curve analysis based on the risk score was used to evaluate the predictive value of the six micNETs and six TFs for the mesenchymal subtype in TCGA Agilent and HiSeq databases. (E-F) The pie plot shows that mesenchymal samples predicted by micNETs were mainly consistent with the mesenchymal subtype reported by Verhaak et al. in TCGA Agilent and HiSeq databases.

Figure 5

GBM patients with low risk scores had a greater benefit from TMZ chemotherapy than GBM patients with high risk scores. (A-D) Patients with low micNET risk scores received a significant benefit from TMZ treatment in TCGA Agilent database, TCGA RNA-seq database, TCGA u133a database and CGGA RNA-seq database. (E-H) Patients with high micNET risk scores showed a poorer treatment effect than patients with low risk scores in the same database.

Figure 6

TGFBR2 promotes the mesenchymal phenotype in GBM cells. (A) micNETs with connection degrees of greater than 200 were subjected to ClueGO analysis, and major mesenchymal-related signatures are presented. (B) The locations of six micNETs were predicted by the COMPARTMENTS subcellular localization database. (C) Mesenchymal subtype-related genes were selected according to the report by Carro et al. in 2011. (D) U87, LN229 and U251 cells were transfected with the CDS and 3'-UTR of TGFBR2 (2.5 µg/well in a six-well plate) for 48 h. U87 and LN229 cells were transfected with si-TGFBR2 (50 nmol/mL) for 48 h. Mesenchymal subtype-related signature genes were measured by real-time PCR.

Figure 7

The TGF-β pathway could be a therapeutic target for the mesenchymal subtype of GBM. (A) Operations were performed on patients with diagnosed GBM. Tumors were obtained and evaluated by histopathological examination. (B) Tumor tissues were cut into 1-mm³ blocks and separated into two groups. One group was treated with DMSO, whereas the other group was treated with LY2109761, an inhibitor of TGFBR2. RNA was extracted from the tissue blocks, and mesenchymal-related signature genes were evaluated by real-time PCR. (C) The tissue blocks of TBD0207 and TBD0220 were subjected to RNA-seq and miRNA-seq. (D) The differentially expressed genes (FDR < 0.001, log2FC > 1) were subjected to GO and KEGG pathway analyses. (E) PDX models were established to test the efficiency of LY2109761 in vivo. (F) Immunohistochemistry was performed to detect Ki-67, TGF-β and CD34 expression. (G) Tumors derived from the PDX model were tested by RNA-seq analysis. (H) Mfuzz was performed to detect alternative expression profiles.

Figure 8

Scheme of mesenchymal GBM subtype progression.