Pathway-based approach reveals differential sensitivity to E2F1 inhibition in glioblastoma

Abstract

Analysis of tumor gene expression is an important approach for the classification and identification of therapeutic vulnerabilities. However, targeting glioblastoma (GBM) based on molecular subtyping has not yet translated into successful therapies. Here, we present an integrative approach based on molecular pathways to expose new potentially actionable targets. We used gene set enrichment analysis (GSEA) to conduct an unsupervised clustering analysis to condense the gene expression data from bulk patient samples and patient-derived gliomasphere lines into new gene signatures. We identified key targets that are predicted to be differentially activated between tumors and were functionally validated in a library of gliomasphere cultures. Resultant cluster-specific gene signatures associated not only with hallmarks of cell cycle and stemness gene expression, but also with cell-type specific markers and different cellular states of GBM. Several upstream regulators, such as PIK3R1 and EBF1 were differentially enriched in cells bearing stem cell like signatures and bear further investigation. We identified the transcription factor E2F1 as a key regulator of tumor cell proliferation and self-renewal in only a subset of gliomasphere cultures predicted to be E2F1 signaling dependent. Our in vivo work also validated the functional significance of E2F1 in tumor formation capacity in the predicted samples. E2F1 inhibition also differentially sensitized E2F1-dependent gliomasphere cultures to radiation treatment. Our findings indicate that this novel approach exploring cancer pathways highlights key therapeutic vulnerabilities for targeting GBM.

Keywords: CIP2A; E2F1; Glioblastoma; Molecular targets; Tumoral heterogeneity.

FIGURE 1

Pathway-based analysis generates three distinct clusters based on enrichment profiles with clinical significance. Samples from TCGA were analyzed using either canonical pathways (A) or oncogenic pathways (B) from the GSEA to generate heatmaps based on the enrichment profile of each sample (column) with respect to each gene set (row) in both collections. C and D, Profiles from A and B, respectively, were used to generate PCA plots labeled by color and shape for each cluster. Circle lines represent the normal distribution of the samples in each cluster. E, TCGA samples were clustered on the basis of the original molecular subtypes described, and Kaplan–Meier curves were obtained. F and G, Samples clustered on the basis of enrichment profiles for canonical and oncogenic gene sets, respectively, were analyzed for survival using Kaplan–Meier curves. Tables at the bottom describe the distribution of the molecular subtypes for each cluster. Dotted lines represent median survival for each curve (also described in top tables). Time shown is in months. P values after post hoc analyses using Bonferroni–Hochberg correction.

FIGURE 2

Gene lists predict E2F1 as a main target in one of the clusters found in the GS dataset. A, Enrichment profiles using gene lists were generated for GS samples. B, Each gene list was evaluated using IPA and top predicted activated (green arrows) and inhibited (red arrows) upstream regulators are shown. C, PCA plot from enrichment scores generated in A showing how each gene list contributes to a particular direction. D, Samples from both clusters were evaluated for their enrichment of cell cycle–related, downstream E2F1 target, and DNA damage repair gene sets from the canonical pathway collection. E, WGCNA generated 26 modules when samples were analyzed on the basis of their enrichment profiles for the gene lists. F, Modules are ranked on the basis of their abundance in both clusters. Modules at the top are highly enriched in the E2F1-activated cluster (cluster 1).

FIGURE 3

Gene lists differentially correlate with cellular states and cell-specific markers. A, Scores generated for each cell in Supplementary Fig. S3 using gene lists were correlated with scores for cellular states and specific cell-type markers in development and adult brain. The presence of a circle represents significant correlation, and the size and color depict the intensity of the correlation. Boxes mark groups of gene lists strongly correlated (based on hierarchical clustering).

FIGURE 4

E2F1 silencing compromises self-renewal and proliferation in vitro and tumor formation in vivo. Samples from both clusters were treated with control (scrambled) or E2F1 siRNA (A) or were plated in regular media or media containing fulvestrant or calcitriol (B) under limiting dilution in a 96-well plate. Graphs depict the number of wells that did not form spheres after 10 days versus the number of cells plated (a vertical line implies all wells formed spheres). C, Cells treated with scrambled or E2F1 siRNA were plated at a density of 2,000 cells per well in a 96-well plate in quadruplicate, and their growth was evaluated using luminescence. Relative growth is the fold change compared with basal measurement. D–F, Cells treated with scrambled or E2F1 shRNA were intracranially injected in NSG mice. Kaplan–Meier survival curves for each group was calculated; dashed lines represent median survival and time shown is in weeks (D). Luminescence was assessed 2 weeks after transplantation (E). Quantification for each group is shown at 2, 5, and 8 weeks (F). Mice in both groups for HK408 did not reach the 8-week timepoint. Experiments in A–C were performed at least three times. Data are represented as mean ± SEM. **, P < 0.01 and ***, P < 0.001 as assessed by one-way ANOVA.

FIGURE 5

E2F1 silencing compromises DNA damage response induced after irradiation. A, Cells were treated with either C (control) or E (E2F1) shRNA, were subjected to irradiation (8 Gy) and fixed after 12 hours for γH2AX staining (red). Nuclei were counterstained using DAPI. B, Quantification for each group in A is shown. Experiment was performed at least two times. Data are represented as mean ± SEM. **, P < 0.01 as assessed by one-way ANOVA.