Comprehensive analysis of PD-L1 expression in glioblastoma multiforme

Abstract

Glioblastoma multiforme are highly malignant brain tumours with frequent genetic and epigenetic alterations. The poor clinical outcome of these tumours necessitates the development of new treatment options. Immunotherapies for glioblastoma multiforme including PD1/PD-L1 inhibition are currently tested in ongoing clinical trials. The purpose of this study was to investigate the molecular background of PD-L1 expression in glioblastoma multiforme and to find associated pathway activation and genetic alterations. We show that PD-L1 is up-regulated in IDH1/2 wildtype glioblastoma multiforme compared to lower-grade gliomas. In addition, a strong association of PD-L1 with the mesenchymal expression subgroup was observed. Consistent with that, NF1 mutation and corresponding activation of the MAPK pathway was strongly connected to PD-L1 expression. Our findings may explain different response to PD-L1 inhibition of patients in ongoing trials and may help to select patients that may profit of immunotherapy in the future.

Keywords: PD-L1; WGCNA; glioblastoma multiforme; immunotherapy; integrative analysis.

Figure 1

A. A boxplot PD-L1 expression in patients with WHO grad II, III and glioblastoma multiforme (WHO grade IV). B. A boxplot of PD-L1 expression in all expression subclasses define by Verhaak et al., 2010 of high-grade gliomas (glioblastoma multiforme). C-D. Survival Analysis of the TCGA database and Freiburg cohort. PD-L1 high vs low was determinate by the mean expression +/− standard deviation. Patients of the Freiburg cohort with outstanding events were censored and marked by a cross. *** p<0.001, ** p=0.01, *p<0.05.

Figure 2

A. A boxplot of PD-L1 expression in lower-grade gliomas and its molecular subgroups. B. A boxplot of PD-L1 expression in IDH mutated and non-mutated glioblastoma multiforme. D. A boxplot of PD-L1 promoter methylation in IDH1/2 wild-type and mutated patients. C. Functional analysis of methylation/expression correlation of different CpG sides of the PD-L1 promoter. The strongest negative correlation was found in the cg19724470. Significant values were corrected by false-discover rate (FDR). (D) This CpG side was used to validate the mean methylation of IDH mutated vs. non-mutated patients. *** p<0.001, ** p=0.01, *p<0.05.

Figure 3

A. Co-mutation plot of highly frequent mutations in glioblastoma multiforme. Mutational frequencies were given in a bar plot beside the co-mutation plot. General mutation load as total numbers of mutations in each patient was presented in the top barplot. PD-L1 expression was illustrated in a barplot in the panel below. Patients with missing expression values were marked by *. B. Boxplots of PD-L1 expression (TCGA) in NF1 wild-type and mutated/deleted samples were given in the left panel. The validation cohort of Freiburg patients showed similar expression differences (right panel). C. Protein-level of p-c-RAF and p-S6 in mutated/deleted and wild- type samples were illustrated. D. Weighted Proteome Network Analysis of the proteome data investigated protein module 1 as highly correlated to PD-L1 expression level. E. Created network analysis of protein module 1 based on intramodule connectivity, derived from WPNA.

Figure 4

A-C. Immunostaining of PD-L1 and phospho-MAPK14 in two patients with NF1 deletion (A) and two patient without NF1 alteration (C). Additionally, six independent fields were quantified by mean signal intensity by ImageJ, normalized and illustrated in a boxplot. D. Correlation analysis of protein-level of PD-L1 and p-MAPK showed a strong positive correlation (r=0.6 p<0.01). E-F. Immunostaining of phospho-JNK and phospho-AKT in a patient with NF1 deletion (upper panel) and one patient without NF1 alteration (lower panel). Additionally, quantification was given in the right boxplot (F). *** p<0.001, ** p=0.01, *p<0.05.

Figure 5

A. Weighted gene co-expression network analysis of the whole transcriptome data. B. A cluster of clusters analysis presented connected expression modules. Expression module 1 and 2 were contained in one cluster. C. Detailed cluster branches of expression module 1 and 2 were illustrated. Corresponding correlation heatmaps of PD-L1 expression and module-contained genes are given in the panel below the branches. Scatterplots of intramodule connectivity (KME) confirmed the strong correlation of PD-L1 expression and the expression modules. D. Gene Set Enrichment Analysis of expression module 1 identified INFγ related genes (upper panel) and immune response (bottom panel). E. Created network analysis of expression module 1/2 based on intramodule connectivity, derived from WGCNA. Size and colours indicate the intensity of intramodule connectivity.