TCF4and STAT3Transcription By ChIP-seq And Glioblastoma Molecular Subclassification

Glioma is the most common primary brain tumors. Malignant gliomas including WHO III and IV grade gliomas, accounts for80%of all gliomas. Glioblastoma multiforme (GBM), the most malignant subtype, is infiltrative and aggressive, with median survival approximately12months after diagnosis in the majority of cases, despite advances in surgical resection, radiation, and chemotherapy. Recent progress in the research of the development of glioma has been made, however, its real initial etiology remains poorly understood and warrants further investigation. Now widely used WHO classification based on macro analysis of histopathology, cannot exactly reflect glioma cell biological behavior and the sensitivity to radiotherapy and chemotherapy, and is unable to propose effective therapy targets. Cancer molecular classification is developing towards to the new diagnosis system based on molecular pathology signature for predicting the response to therapy, the survival and recurrence and applying to individualized comprehensive treatment. Therefore, it is essential to investigate the mechanism involved in the development and progression and molecular classification of glioma, to explore more effective diagnosis and therapeutic strategies.Wnt/β-catenin signaling and signal transducers and activators of transcription3(STAT3) signaling are crucial factors in the development of glioma. They form the transcriptional complex in the nucleus and regulating the expression of multiple genes. β-catenin/TCF4complex is the core transcriptional complex of Wnt/β-catenin signaling, and the dimerization of phosphorylated STAT3is the core transcriptional complex of STAT3signaling. Comparing to normal brain tissues, the level of P-catenin and STAT3is positively correlated with glioma grade. Repression of β-catenin/TCF4or STAT3transcriptional activity respectively, reduces glioma cell growth and invasion, induces cell apoptosis, and inhibits tumor growth in xenograft nude mouse model. Recent data have shown that β-catenin/TCF4and STAT3have the complex reciprocal regulation network, including interactive and synergetic transcriptional regulation. Thus, the relationship between β-catenin/TCF4and STAT3 in gliomal and its clinical significance are required to be further investigated. In our present study, we aim to investigate genome-wide screening of β-catenin/TCF4and STAT3binding characteristics and co-regulated target genes by TCF4and STAT3in GBM by performing chromatin immunoprecipitation followed by sequencing (CHIP-Seq). Then, we will explore the guiding role of TCF4and STAT3target genes in GBM molecular classification by integrating the gene expression profiles of GBM of Western and Asian population, to propose a novel GBM molecular classification with meaningful clinical significance. Finally, we will investigate the association of EZH2(one of TCF4and STAT3common target genes) with glioma grade and survival, identify EZH2targets and associated genes, and analyze the regulation association of EZH2with TCF4and STAT3.In the first part of our study, to identify genes regulated by TCF4and STAT3at genome-wide level, we performed ChIP-Seq upon human GBM U87cells using TCF4-and STAT3-specific antibodies, respectively. After removing duplication caused by PCR amplification, we obtained a total of more than10million uniquely mapped non-duplicate reads (10.5million for TCF4,9.1million for STAT3). To determine the binding sites, we used a poisson-based method (MACS) upon the uniquely mapped non-duplicate reads with the use of P value threshold of0.001and m-fold value adjusted to be able to establish calculation models. We obtained8307binding sites for TCF4and6908binding sites for STAT3. The binding sites for TCF4and STAT3had similar chromosome distribution, and mainly located at chromosome1,2,3,4,7. To comprehensively understand the genomic distribution of binding sites, we associated each binding site with a known RefSeq gene (+/-50kb of coding region). Given a gene, its promoter region was defined as+/-2kb of its TSS (transcriptional start site). The long-distance regulatory regions of a gene incorporated genomic areas from50kb to2kb upstream of its TSS and from its TES (transcription end site) to50kb downstream of its TES. We discovered that the genomic distribution of TCF4-and STAT3-binding sites were similar. Approximately63%binding sites located in+/-50kb of gene coding regions (3%in promoter,2%in exon,36%in intron,22%in long-distance regulatory regions for TCF4; and3%in promoter,2%in exon,37%in intron,21%in long-distance regulatory regions for STAT3). Notably, the substantial proportion of binding sites was identified in long-distance regulatory regions, suggesting that TCF4and STAT3also be able to regulate genes in sites far away from the proximal promoter regions. The most enriched motifs identified for TCF4and STAT3were ATCATACTAT and ATCAACATAT, respectively. By associating each binding site with the nearest gene within50kb upstream and50kb downstream of coding region, we identified a total of3812TCF4target genes and3165STAT3target genes. There were overlapped genes (1250genes) targeted by both TCF4and STAT3.To understand the biological processes co-regulated by TCF4and STAT3, we performed gene ontology enrichment analysis upon their co-targeted genes using DAVID with the p value threshold of0.05. The genes targeted by both TCF4and STAT3were enriched in developmental processes of nervous system including neuron differentiation, neuron migration, neuron projection morphogenesis, cell morphogenesis involved in neuron differentiation and neuron development.In the second part of our study, we overlapped1250TCF4and STAT3target genes with a large cohort of gene expression profiling ftom TCGA database consisting of173core GBMs.801out of1250TCF4and STAT3co-targeted genes were covered in the TCGA cohort. We then performed hierarchical clustering upon the173GBMs with these801co-targeted genes, revealing that there were four subtypes for GBMs, with a larger degree of agreement in assignment of samples to TCGA subtypes (Classical, Mesenchymal, Neural and Proneural), suggesting that TCF4and STAT3could co-guide TCGA sub-classification of GBMs. Then we obtained a set of132co-targeted genes by selecting the most differentially expressed genes among our GBM subtypes by performing pair-wise comparison with the use of SAM. The hierarchical clustering of GBMs in TCGA with the132co-targeted genes resulted in four tumor subtypes, consistent with TCGA GBM subtypes. Functional annotation clustering by DAVID of the132co-targeted genes demonstrated that they were highly enriched in developmental processes of nervous system. Therefore, we test whether genes associated with nervous system development can classify GBMs into different subtypes. Top100genes statistically enriched in developing astrocytes and top100genes statistically enriched in oligodendrocyte progenitor cells were respectively extracted from the brain transcriptome database, followed by using them to perform hierarchical clustering upon GBMs in TCGA cohort. The GBMs in TCGA cohort were clustered into four subtypes, also largely consistent with TCGA GBM subtypes. Based on aggressive treatment efficacy and clustering signature, we proposed a novel GBM classification:GBM can be divided into two major types, Mesenchymal-like type (containing Classical and Mesenchymal subtypes) and Proneural-like type (containing Neural and Proneural subtypes). To explore the gene signature of these two new subtypes, a set of142co-targeted and most differentially expressed genes between Mesenchymal-like and Proneural-like types were obtained using SAM form801co-targeted genes. Hierarchical clustering with these142genes classified173GBMs into two subtypes, Mesenchymal-like and Proneural-like types, with a high agreement. Survival analysis of109GBM patients with treatment data showed that there was no statistically significance between these two subtypes. Further, we examined the effect of temozolomide (TMZ) treatment, defined as concurrent TMZ chemotherapy and radiotherapy. Response to TMZ therapy differs by subtype, with the benefit in Mesenchymal-like subtype and no benefit in Proneural-like subtype. These findings indicate an important contribution toward the ability of142-gene classification to study GBM subtypes, especially for modeling and predicting therapeutic response. An independent set of260GBM expression profiles (also downloaded from TCGA database) was compiled from the public domain to assess subtype reproducibility. Applying a same ordering of gene list and hierarchical clustering in the validation set clearly recapitulated the gene sample groups. Given that TCGA data and validation data were generated from Western population, the existence of Mesenchymal-like and Proneural-like subtypes in Asian population is unclear. Therefore, we used220glioma samples from Chinese patients in an attempt to complement and validate our molecular subtyping system with142genes.141out of142co-targeted genes were covered in the Chinese cohort. Hierarchical clustering in the Chinese validation set clearly represented Mesenchymal-like and Proneural-like subtypes, with the same ordering of gene list. Then we focused on survival analysis of GBM Mesenchymal-like and Proneural-like subtypes. It did not reach statistically significance comparing the survival data in GBM Mesenchymal-like and Proneural-like subtypes. Further, the treatment efficacy of TMZ was evaluated in GBM patients. Mesenchymal-like subtype predicted significantly better outcome at the status of TMZ treatment. However, TMZ treatment did not alter survival in the Proneural-like subtype. These data were consistent with the survival characteristics of Mesenchymal-like and Proneural-like subtypes of TCGA GBM samples, suggesting that GBM Mesenchymal-like and Proneural-like subtypes based on142-gene classification could be used for modeling and predicting therapeutic response in Western population and Asian population.In the third part of our study, we found that, EZH2, as one of TCF4and STAT3target genes, was positively associated with glioma grade. And the increased level of EZH2expression in GBMs confers poorer overall survival. Univariate COX regression analysis showed that EZH2was an independent prognostic factor in GBM patients. The expression of EZH2is positively associated with SUZ12and EED, two other components in Polycomb Repressive Complex2(PRC2). By integrating other data of EZH2targets, we identified312EZH2targets in GBMs. The expression mode of EZH2targets in GBMs with lower EZH2expression became closer to that in normal brain. Functional annotation analysis of312EZH2targets demonstrated that they were highly enriched in developmental processes of nervous system. Further, we screened228EZH2associated genes in GBMs by correlation analysis. The expression mode of EZH2associated genes in GBMs with lower EZH2expression became closer to that in normal brain. Functional annotation analysis of EZH2associated genes showed that they were highly enriched in cell cycle, nuclear division and DNA replication. Repression EZH2by RNAi induced glioma cell cycle G0/G1arrest. Knock down EZH2using RNAi delayed tumor growth in nude mice harboring subcutaneous U87xenografts. Finally, inhibition β-catenin/TCF4and STAT3transcriptional activity by small molecular inhibitors FH535and WP1066respectively triggered a reduction of EZH2expression. And downregulation of EZH2decreased the level of β-catenin and STAT3.Similar to results obtained from in vitro analyses, the expressions of β-catenin and STAT3were significantly reduced in tumor specimens from the treatment group.In conclusion, it is the first comprehensive study to investigate TCF4and STAT3transcription at genome wide level by Chip-Seq and data mining of patient cohorts to implicate with molecular subclassification of GBM.(1) TCF4and STAT3regulate many genes through long-distance regulatory regions in GBM cells, except for the promoter regions. TCF4and STAT3have similar transcription feature and synergeticly modulate the downstream targets. These will provide a powerful approach for identifying potential gene signatures with biological and clinical importance and lead to new insights in GBM tumorigenesis.(2) To our knowledge, this is the first study to link TCF4and STAT3co-regulated genes with developmental genes in nervous system for molecular subclassification of GBMs. Of note, we propose a novel classification of GBM into two major types, Proneural-like and Mesenchymal-like types, which in Western and Asian population, indicating a contribution towards to guiding for individual treatment.(3) EZH is associated with glioma grade and prognosis, and plays an important role in the development of glioma. We identify a cohort of EZH2targets and associated genes and reveal pasitvie feedback network among EZH2, β-catenin/TCF4and STAT3. These findings provide a potential approach for identifying diagnostic and prognostic biomarkers and therapy targets, and lighting the tumor tumorigenesis and therapy strategy optimization in glioma.