Glioblastoma evolution pattern under surgery and radio(chemo)therapy (RCHT) to identify novel methylome based glioma subtypes.

Abstract

2012 Background: Identification of isocitrate dehydrogenase mutations (IDHm) and glioma CpG island methylator phenotype (G-CIMP) as well as methylation of O6-methylguanine DNA methyltransferase (MGMT) promotor has substantially improved stratification of glioma patients into prognostic subgroups. In extension of static pre-therapy diagnostic, we sought to investigate the impact of glioblastoma evolution under selection pressure of standard therapy on methylome level. Methods: For the training cohort (T), methylome (450k Illumina) data of paired samples from 50 patients with glioblastoma (GBM, 11 G-CIMP+) were analyzed, i.e., primary (P) and at the time point of recurrence (R, re-surgery) after standard therapy at NCT. For 39 pairs matching RNASeq data was analyzed. Validation cohorts consisted of Heidelberg (V1) total n = 650, GBM (n:585, 8 G-CIMP+), grade III (n:65, CIMP+ 65), Austrian GBM (V2, n = 499, 36 IDHm, pyrosequencing data) and the TCGA (V3) Lower-grade-glioma cohort (LGG, n = 477, grade III n: 247, 178 G-CIMP+, grade II n: 228, 206 G-CIMP+). Results: Limited number of consensus differentially methylated probes (DMP) were found across all P vs. R samples (nCpG = 411 CpGs, FDR < 0.05). In contrast, heterogeneity in GBM evolution was found by similarity analysis of delta-methylome data of 50 PR pairs resulting in two distinct clinical subgroups and one “intermediate” group. Intriguingly, n = 114.652 DMP (FDR < 0.05) was found by comparing the evolutionary “poor” (n = 15) vs. “good” (n = 13) GBM phenotypes. A random forest classifier was built to identify the evolutionary subgroups in P samples. The performance of “good” prognosis classifier was in T cohort HR: 0.54 [0.30-0.97], p = 0.04; V1: 0.57 [0.43-0.76], p < 0.001, V2: 0.62 [0.47-0.82], p < 0.001, LGG: 0.16 [0.08-0.32], p < 0.001. In “good” prognosis group (T), neither G-CIMP+ (n = 3) nor MGMT-STP27 (oddsratio, OR: 0.56, p = 0.47) was enriched. MGMT-STP27 OR was 0.47 (V1, p = 0.47) or 1.28 (V2, p = 0.45), respectively. The evolutionary subgroups remain prognostic independent of GCIMP status in LGG (V3). “Poor” glioma are enriched for RTKI/II methylome subtypes, and contain less frequently the mesenchymal subtype. Bevacizumab treatment showed a survival benefits only in “poor” subtype (V1+2). Conclusions: Discovery of a methylome based classifier of glioma evolution informs on “good” and “poor” prognosis subtypes and may have ramification for stratifying patients for therapy such as e.g., antiangiogenesis.