Multidimensional analysis of matched primary and recurrent glioblastoma identifies contributors to tumor recurrence influencing time to relapse

Abstract

Glioblastoma (GBM) is a lethal brain tumor without effective treatment options. This study aimed to characterize longitudinal tumor changes in order to find potentially actionable targets to prevent GBM relapse. We extracted RNA and proteins from fresh frozen tumor samples from patient-matched IDHwt WHO grade 4 primary (pGBM) and recurrent (rGBM) tumors for transcriptomics and proteomics analysis. A tissue microarray containing paired tumor samples was processed for spatial transcriptomics analysis. Differentially expressed genes and proteins between pGBM and rGBM were involved in synapse development and myelination. By categorizing patients into short (STTR) and long (LTTR) time-to-lapse, we identified genes/proteins whose expression levels positively or negatively correlated with TTR. In rGBM, expressions of Fcγ receptors (FCGRs) and complement system genes were negatively correlated with TTR, whereas expression of genes involved in DNA methylation was positively correlated with TTR. Spatial transcriptomics of the tumor cells showed enrichment of oligodendrocytes in rGBM. Besides, we observed changes in the myeloid compartment such as a switch from quiescent to activated microglia and an enrichment in B and T cells in rGBM with STTR. Our results uncover a role for activated microglia/macrophages in GBM recurrence and suggest that interfering with these cells may hinder GBM relapse.

Keywords: glioblastoma recurrence; microglia transcriptomic; proteomic; spatial transcriptomic.

Figure 1.

Targeted transcriptomic analysis of patient-matched pGBM and rGBM reveals synapse formation and oligodendrocyte differentiation as pathways activated in rGBM. (A) Schematic experimental setup for gene expression analysis using RNA extracted from fresh-frozen samples from 15 patient-matched pGBM and rGBM tumors using Nanostring nCounter system. (B) PCA plot of all samples included in the analysis, each dot represents a patient, color-coded according to pGBM or rGBM. Dim1 represents PC1 and Dim2 PC2. The percentage of variation explained by each component is shown in parentheses. (C) Hierarchical clustering heatmap of the genes expressed across all samples. Z-score transformation was performed for each protein. Genes with similar expression patterns cluster together. Relative expressions were scaled from red (high expression) to blue (low expression). Each column represents a sample (dark green: pGBM, light green: rGBM), and each row represents a gene. (D) Hierarchical clustering heatmap of the differentially expressed genes across all samples. (E) Volcano plot showing significantly differentially overexpressed genes in rGBM (light green) and pGBM (dark green) (adjusted P < .05, log₂FC > 0 and log₂FC < 0). (F, G) Dot plots of the enriched biological processes (GO-BP) determined by ORA using overexpressed genes in rGBM and pGBM, respectively.

Figure 2.

Unbiased LC-MS based proteomic confirms synapse signaling as a major pathway in rGBM. (A) Experimental setup for LC-MS based proteomic analysis of 6 fresh frozen patient-matched pGBM and rGBM samples (clinical parameters highlighted in Table 1 and Table S1). (B) PCA plot of all samples included in the analysis, each dot represents a patient, color coded according to pGBM or rGBM. Dim1 represents PC1 and Dim2 PC2. The percentage of variation explained by each component is shown in parentheses. (C) Hierarchical clustering heatmap of all the identified proteins across all samples. Z-score transformation was performed for each protein. Proteins with similar expression patterns were clustered together. Relative expressions were scaled from red (high expression) to blue (low expression). Each column represents a sample (dark green: pGBM, light green: rGBM), and each row represents a protein. (D) Hierarchical clustering heatmap of all DEPs across all samples. (E) Volcano plots showing significantly differentially overexpressed proteins in rGBM (light green) and pGBM (dark green) (adjusted P < .05, log₂FC > 0 and log₂FC < 0). (F, G) Dot plots of the enriched biological processes (GO-BP) determined by ORA using overexpressed proteins in rGBM and pGBM, respectively.

Figure 3.

Expressions of immunoglobulin binding receptors and complement system components are higher in the subset of patients with STTR and inversely correlate with TTR. (A) Kaplan-Meier Estimates of time to relapse in rGBM for patients with STTR (<309 days) and LTTR (≥309 days). (B) Volcano plot showing DEGs between patients with STTR vs LTTR in rGBM samples. Genes overrepresented in rGBM with STTR (purple) or LTTR (orange) are highlighted (adj. P < .25, logFC > 0.5 and logFC < −0.5), dot-plots of the corresponding enriched biological processes (GO-BP) determined by ORA (right part) and expression of genes enriched in at least one of the 4 most enriched processes (lower part). (C) Scatterplots showing expression of the top negatively (r < −0.63, left) or positively (r ≥ 0.63, right) correlated gene with TTR in rGBM samples.

Figure 4.

Spatial transcriptomic of CD64- and GFAP-expressing cell populations revealed subtle changes in microglia subtypes in rGBM. (A) Representative tissue microarray section containing 7 pairs of patient-matched pGBM and rGBM samples stained for GFAP (red), CD64 (green), and DAPI (cell nuclei in blue) identifying ROIs for GFAP+, CD64+ cellular fractions and nuclei, respectively, and selected for whole transcriptome sequencing. Inserts show magnifications of ROIs from pGBM and rGBM from patient BTB38 as an example. (B) PCA plot for CD64+ samples in pGBM and rGBM. Dim1 represents PC1 and Dim2 PC2. The percentage of variation explained by each component is shown in parentheses. (C) Volcano plot showing DEG between pGBM and rGBM in CD64+ cells. (D) Volcano plot showing DEG between STTR and LTTR in CD641 rGBM cells. (E) PCA plot for GFAP+ samples in pGBM and rGBM. Dim1 represents PC1 and Dim2 PC2. The percentage of variation explained by each component is shown in parentheses. (F) Volcano plot showing DEG between pGBM and rGBM in GFAP+ cells. (G) Volcano plot showing DEG between STTR and LTTR in GFAP+ rGBM cells. For all the volcano plots P < .05 in light blue, adj. P < .05 in dark blue, vertical dashed lines at logFC > 0.5 or < −0.5, horizontal dashed line at P < .05.

Figure 5.

Cellular deconvolution of CD64+ and GFAP+ cell populations revealed subtle changes in microglia subtypes in rGBM. (A) Relative cell type abundance in the CD64+ population, divided into pGBM and rGBM samples (left chart) and grouped according to short or long TTR (right chart). (B) Relative cell type abundance in the GFAP+ population was divided into pGBM and rGBM samples (left chart) and grouped according to short or long TTR (right chart).