Transcriptome-wide DNA Methylation Alterations In Glioblastoma

Background: Glioblastoma multiforme(GBM)presented as molecularly heterogeneous groups of diseases with invariably dismal prognosis.Among all molecular machineries,DNA methylation played crucial roles in determining tumor biological behaviors via modulating gene expression and chromatin structure.GBMs had been known for their distinct epigenetic landscapes featured by global DNA hypomethylation in gene-poor DNA repeats,and large hypomethylated blocks concurrently with mosaicked hypermethylation of CpGs islands(CGI)in gene regions.Unfortunately,limited data had been acquired on the roles of cancer-specific DNA hypomethylation.It had been reported that large-scale hypomethylation of gene-poor DNA repeats was mainly responsible for global DNA hypomethylation in cancers.Cancer-specific DNA hypermethylation mainly came from gene promoter-associated CGI.However,despite that nearly half of all human gene promoters do not contain CGIs;the relevance of DNA methylation changes at non-CGI gene regions remained largely elusive in cancers.This study was aimed to investigate genome-wide DNA methylation alterations in GBMs,with a focus on non-CGI open sea regions.Methods: Totally 129 newly diagnosed GBMs(≥18 years old)and 6 non-tumor brain tissues were collected between 2004-2013 from Rennes and Angers University Hospitals(RAUH).Samples were collected,following informed consent,in accordance with the French regulations and the Helsinki Declaration.All GBMs were treated with radiotherapy(RT)with concurrent and adjuvant temozolomide(TMZ).Infinium HumanMethylation27 k or 450 k BeadChip and Agilent Whole Genome 4×44K Microarray Kit were applied for genome-wide DNA methylation and gene expression profiling respectively.Molecular datasets of all gliomas and non-tumor brains were also obtained from The Cancer Genome Atlas(TCGA)and the Gene Expression Omnibus(GEO seires.CpGs probes that were associated with any genes(protein-coding or non-protein-coding)were included for analysis.Differentially methylated CpGs and differentially expressed genes were calculated by wilcoxon test and standard t test.GBM-specific CpGs were defined as those having a median β difference ≥ 0.2 between tumors and controls,and a false discovery rate(FDR)q-value ≤ 0.05.Correlation of DNA methylation and gene expression was assessed by Pearson correlation analysis.Methylation-dependent expression was defined as absolute Pearson r coefficient ≥ 0.3 and P value ≤ 0.05.The discovery-validation approach was employed for risk-score model.Bioinformatic analyses were conducted by Gene Set Enrichment Analysis(GSEA)and DAVID funcational annotation clustering.Principal components analysis(PCA)was carried out to assess DNA methylation patterns of gliomas of each grade and histological type.Differences in clinical or molecular features within grouped samples were tested by unpaired t test,Fisher’s exact or Chi-square test.Overall survival(OS)was estimated by the Kaplan-Meier method,and compared by log-rank test.All calculations were done within R software,with two-sided p ≤ 0.05 for significance.Results: We totally identified a panel of 801 CpGs(399 hypermethylated and 402 hypomethylated)that were commonly identified in at least three of the four analyzed datasets as a representative differential list for GBMs.PCA supported the robustness of the list as differential epigenetic alterations for GBMs.Methylation changes of the 801-CpGs signature showed distinctive distributions with regard to their epigenomic locations: nearly all CpGs from CGIs(97%)gained DNA methylation in GBMs,and the proportion was decreased in shores(75%)and the flanking shelves(37%),whilst the bulk of open sea loci(93%)lost DNA methylation in GBMs.We found that genes relevant to open seas were associated with lower frequency of differential expression status in GBMs as compared to those relevant to CGIs and the flanking regions.Similarly,open sea loci were associated with lower frequency of methylation-dependent expression.Upon correlation with OS,we found that DNA methylation of open sea CpGs(96-100%)was nearly all in positive correlation with OS whilst that of CGI CpGs(96-100%)were almost exclusively in inverse correlation with GBM prognosis.DAVID analysis reported that,among the 801-CpGs signature,CGI loci mostly affected genes of developmental and transcription processes whilst open seas favored genes involved in immune-related processes.Based on risk-score model,we constructed a panel of 8 open sea CpGs relating to an immune-relevant gene signature from two discovery cohorts with RT/TMZ.The risk-score model was constructed as follows: risk score =(-1.07 ×β value of cg02833180)+(-3.06 ×β value of cg09238677)+(-1.89×β value of cg18343292)+(-1.54×β value of cg18854666)+(-1.65×β value of cg18881723)+(-3.44×β value of cg25511807)+(-1.93×β value of cg26066361)+(-1.65×β value of cg22502502).The optimal cutoff was calculated around the 40 th percentile risk score(-5.17)from the discovery cohorts.The 8-CpGs signature could predict OS in two discovery cohorts and four validation cohorts.Univariate and multivariate Cox regression analyses in RAUH samples showed that the 8-CpGs signature was significantly correlated with OS,independent of patient age and MGMT promoter methylation status.GSEA on transcriptome data revealed distinctive local immune profiles between the risk subgroups: high-risk tumors were highly enriched with immune gene sets,indicating an intense local immune response in those GBMs.The prediction of differential transcriptionally immune status by the local epigenetic signature was also validated in independent datasets.Comparative analysis of the GBM-derived 801-CpGs differential methylation signature in lower-grade gliomas(LGGs)showed that GBM-specific CpGs including those from open seas were mostly affected within LGGs but with different extents.We found that GBMs and LGGs showed tremendous increase in overall DNA methylation levels in CGIs as compared to non-tumor brains,whilst he differences were relatively subtle among gliomas of each grade or histological type.In contrast,overall DNA methylation levels in open sea regions were not apparently decreased in LGGs and especially grade II gliomas in comparison with non-tumor brains,but were increasingly reduced in gliomas of higher grades,with GBMs having the lowest overall DNA methylation levels in open sea regions.The 8-CpGs immune signature also robustly predicted OS in two LGG cohorts,with the 40 th percentile risk values for each cutoff,and was highly associated with local immune phenotypes within LGGs.Conclusion: This study performed the first genome-wide evaluation on clinical and functional implications of gene open sea hypomethylation in GBMs.We found that like CGI hypermethylation,gene open sea hypomethylation may also be ubiquitous defining molecular events in GBMs,and be potential candidates for molecular diagnosis.We also observed the limited linkage of open sea hypomethylation to altered expression.Most important,we for the first time revealed the common and exclusive correlation between open sea hypomethylation and unfavorable prognosis in GBMs.Gene open sea hypomethylation may represent new promising candidates for cancer prognostic parameters.The 8-CpGs immune risk signature was demonstrated to be a robust and independent prognostic indicator,and may be used as a clinically useful biomarker.The close relationship between the 8-CpGs signature and differential immune profiles indicated a promising role in guding combination of epigenetic and immunotherapy.The different altering timings may indicate the distinct roles of DNA methylation alterations in different epigentic regions: CGI hypermethylation may have major roles in glioma initiation whilst open sea hypomethylation may determine the degree of tumor aggressiveness.In summary,this study provided initial data highlighting the potential value of gene open sea hypomethylation in gliomas;further studies are much needed for fully characterizing these undervalued epigenetic changes in cancer biology.