The DNA methylome of DDR genes and benefit from RT or TMZ in IDH mutant low-grade glioma treated in EORTC 22033

Abstract

The optimal treatment for patients with low-grade glioma (LGG) WHO grade II remains controversial. Overall survival ranges from 2 to over 15 years depending on molecular and clinical factors. Hence, risk-adjusted treatments are required for optimizing outcome and quality of life. We aim at identifying mechanisms and associated molecular markers predictive for benefit from radiotherapy (RT) or temozolomide (TMZ) in LGG patients treated in the randomized phase III trial EORTC 22033. As candidate biomarkers for these genotoxic treatments, we considered the DNA methylome of 410 DNA damage response (DDR) genes. We first identified 62 functionally relevant CpG sites located in the promoters of 24 DDR genes, using the LGG data from The Cancer Genome Atlas. Then we tested their association with outcome [progression-free survival (PFS)] depending on treatment in 120 LGG patients of EORTC 22033, whose tumors were mutant for isocitrate dehydrogenase 1 or 2 (IDHmt), the molecular hallmark of LGG. The results suggested that seven CpGs of four DDR genes may be predictive for longer PFS in one of the treatment arms that comprised MGMT, MLH3, RAD21, and SMC4. Most interestingly, the two CpGs identified for MGMT are the same, previously selected for the MGMT-STP27 score that is used to determine the methylation status of the MGMT gene. This score was higher in the LGG with 1p/19q codeletion, in this and other independent LGG datasets. It was predictive for PFS in the TMZ, but not in the RT arm of EORTC 22033. The results support the hypothesis that a high score predicts benefit from TMZ treatment for patients with IDHmt LGG, regardless of the 1p/19q status. This MGMT methylation score may identify patients who benefit from first-line treatment with TMZ, to defer RT for long-term preservation of cognitive function and quality of life.

Keywords: DDR genes; DNA methylation; Low-grade glioma; MGMT-STP27; Randomized trial; TMZ.

Fig. 1

Heatmap representation of normalized DNA methylation of the 62 functional DDR CpGs in EORTC 22033. The dendrogram was established by hierarchical classification using Ward’s algorithm and Euclidean distance. Annotation of the 132 LGG samples comprises: MGMT methylation score (purple–blue color gradient, higher score more purple), purity estimation (HMP index, orange color gradient, more pure, more dark), copy number status for 10q26.3 region (grey, no change; red, amplification; blue, deletion), 1p/19q codeletion status (green, codeleted; grey, non-codeleted), CIMP status (green, non-CIMP; red, CIMP), IDH1 and IDH2 mutation status (green, wild type; red, IDH mutated; white, not determined), sex (red, female; grey, male), DNA methylation status of MGMT (MGMT-STP27, unmethylated, blue; methylated, red), tissue type (green, frozen; red, FFPE) and Batch (different colors). The 62 functional DDR CpGs are indicated on the right. The corresponding probe names are listed in supplementary Table S2 (Online Resource 2). The location on the chromosome (chr) arm, and the status as differential methylation position (DMP, blue; differentially methylated between codeleted and non-codeleted tumors) are indicated on the left

Fig. 2

Similarity of functional DNA methylation patterns of IDHmt LGG among the four datasets. a The four DNA methylation datasets, using the common 62 functional DDR CpGs, were analyzed simultaneously by the STATIS procedure to determine similarity. First, the global comparison of the four datasets is provided by the interstructure, based on the pairwise comparisons (RV coefficients, all > 0.8). Each dataset is represented by an arrow and the small angles between arrows indicate high similarity between datasets (global permutation test, simulation p value < 0.001 for 999 permutations). b The compromise (PCA-like) analysis of the functional DDR CpGs across the datasets provides an optimized average representation. It gives a general view of the correlations between the functional CpGs (see supplementary Table S2, Online Resource 2, for the list of corresponding CpG probes). The gradient of CpG methylation is indicated with the red arrow. The CpGs detected as significantly differentially methylated between codeleted and non-codeleted groups are represented by blue arrows. The representation of the patients on the two first axes of the compromise analysis is given for each dataset (d, e, f, g). The first axes mainly separates the samples according to the codeletion status, as visualized by the inertia ellipse [CD-II, grade II codeleted (blue); CD-III, grade III codeleted (light blue); N-II, grade II non-codeleted (orange); N-III, grade III non-codeleted (red)]. c In line, the clustering of the molecular subgroups of the four datasets separates the codeleted (blue) from the non-codeleted (red) samples (Ward’s algorithm and Euclidean distance derived from STATIS coordinates and is visualized in a dendrogram)

Fig. 3

DDR networks based on pathways as defined in Pearl et al. [40]. DDR genes identified as functionally methylated are indicated and marked in green (FP). Functional DDR genes that are differentially methylated between IDHmt codel and IDHmt non-codel LGG are marked in red (FDMP)

Fig. 4

Progression-free survival in function of the MGMT methylation score. a The simulation of progression-free survival (%, PFS) is based on the Cox proportional-hazards model and PFS (%), and is shown in function of time (month) and the MGMT methylation score for patients randomized to treatment with radiotherapy (RT) or temozolomide (TMZ). The graphics illustrate a three-dimensional Kaplan–Meier plot, where the MGMT methylation score and time are explanatory variables and the colors of the gradient complemented by contour lines indicate progression-free survival (%). As examples, two cut-offs (first and third quartiles of MGMT score distribution) were used to define three groups of methylation (low, intermediate, high) for patients from EORTC 22033. The two cut-offs are indicated by dashed lines in a. Kaplan–Meier plots illustrate the association between a high MGMT score and PFS in the treatment arms, RT (b) and TMZ (c). The p values are given by log-rank test. The KMs in b and c confirm the significant interaction between TMZ and the MGMT score (Table 2)

Fig. 5

MGMT score in function of the molecular subtype in LGG. The MGMT methylation scores for the three distinct molecular subtypes (wt, CIMP-/IDHwt;  IDHmt, CIMP+ non-codel;  IDHmt codel, CIMP+ codel) separated by WHO tumor grade, are shown for the four datasets (EORTC 22033, TCGA1 & 2, AGlioma). The p values of the comparison between codeleted and non-codeleted samples are based on the Welch’s t test for each dataset. The dashed line indicates the MGMT-STP27 cut-off [logit(0.358)]