Glioblastoma is an aggressive cancer of the brain and spine. While analysis of glioblastoma 'omics data has somewhat improved our understanding of the disease, it has not led to direct improvement in patient survival. Cancer survival is often characterized by differences in gene expression, but the mechanisms that drive these differences are generally unknown. We therefore set out to model the regulatory mechanisms associated with glioblastoma survival. We inferred individual patient gene regulatory networks using data from two different expression platforms from The Cancer Genome Atlas. We performed comparative network analysis between patients with long-and short-term survival. Seven pathways were identified as associated with survival, all of them involved in immune signaling; differential regulation of PD1 signaling was validated to correspond with outcome in an independent dataset from the German Glioma Network. In this pathway, transcriptional repression of genes for which treatment options are available was lost in short-term survivors; this was independent of mutational burden and only weakly associated with T-cell infiltration. Collectively, these results provide a new way to stratify glioblastoma patients that uses network features as biomarkers to predict survival. They also identify new potential therapeutic interventions, underscoring the value of analyzing gene regulatory networks in individual cancer patients. Statement of SignificanceGenome-wide network modeling of individual glioblastomas identifies dysregulation of PD1 signaling in patients with poor prognosis, indicating this approach can be used to understand how gene regulation influences cancer progression.'omics landscape for many different cancer types. Although this has somewhat improved our understanding of the biology underlying the development and progression of cancer(1), it has only led to direct improvement of patient survival for a limited subset of cancer types. For most cancers, Research.

Glioblastoma is an aggressive cancer of the brain and spine. While analysis of glioblastoma `omics data has somewhat improved our understanding of the disease, it has not led to direct improvement in patient survival. Cancer survival is often characterized by differences in expression of particular genes, but the mechanisms that drive these differences are generally unknown. We therefore set out to model the regulatory mechanisms that associate with glioblastoma survival. We inferred individual patient gene regulatory networks using data from two different expression platforms from The Cancer Genome Atlas (n=522 and 431). We performed a comparative network analysis between patients with long- and short-term survival, correcting for patient age, sex, and neoadjuvant treatment status. We identified seven pathways associated with survival, all of which were involved in immune system signaling. Differential regulation of PD1 signaling was validated in an independent dataset from the German Glioma Network (n=70). We found that transcriptional repression of genes in this pathway—for which treatment options are available—was lost in short-term survivors and that this was independent of mutation burden and only weakly associated with T-cell infiltrate. These results provide a new way to stratify glioblastoma patients that uses network features as biomarkers to predict survival, and identify new potential therapeutic interventions, thus underscoring the value of analyzing gene regulatory networks in individual cancer patients.

<p>Differential network topology contrasting the short-term with the long-term survival network. A) Differential network modules significantly enriched for overrepresentation of Gene Ontology terms. Modules are shown for both discovery datasets (D1 and D2). Two modules were significant in both dataset (indicated with M1 and M2). The color in the heatmap indicates the enrichment of genes in the module as observed/expected value (Obs/Exp), with purple representing enriched modules in discovery dataset 1 and green enriched modules in discovery dataset 2. B) "Beeswarm" plot visualizing the distribution of average log differential modularity scores from ALPACA, for transcription factors (TF) and genes (Gene). Higher scores mean higher differential modularity. TFs and genes with significant differential modularity between the two survival groups are labeled.</p>

<p>A) Comparison of distributions of immune scores from xCell between the two survival groups in the discovery and validation datasets. In the boxplots, boxes represent the median and first and third quantiles, whiskers represent 1.5Ã-IQR. B) Association of the immune scores with PD1 targeting scores in the discovery and validation datasets. Regression lines with confidence intervals of 0.95 are shown for the long-term (blue) and short-term (red) groups.</p>

<p>Smooth scatterplot visualizing the correlation between edge weights in the condition-specific network modeled on discovery dataset 1 and the condition-specific network modeled on discovery dataset 2. Pearson correlation coefficient (R) and p-value (p) are indicated in the plot.</p>

<p>Differential network topology contrasting the short-term with the long-term survival network. A) Differential network modules significantly enriched for overrepresentation of Gene Ontology terms. Modules are shown for both discovery datasets (D1 and D2). Two modules were significant in both dataset (indicated with M1 and M2). The color in the heatmap indicates the enrichment of genes in the module as observed/expected value (Obs/Exp), with purple representing enriched modules in discovery dataset 1 and green enriched modules in discovery dataset 2. B) "Beeswarm" plot visualizing the distribution of average log differential modularity scores from ALPACA, for transcription factors (TF) and genes (Gene). Higher scores mean higher differential modularity. TFs and genes with significant differential modularity between the two survival groups are labeled.</p>

<p>A) Comparison of distributions of immune scores from xCell between the two survival groups in the discovery and validation datasets. In the boxplots, boxes represent the median and first and third quantiles, whiskers represent 1.5Ã-IQR. B) Association of the immune scores with PD1 targeting scores in the discovery and validation datasets. Regression lines with confidence intervals of 0.95 are shown for the long-term (blue) and short-term (red) groups.</p>

<p>To ensure that our results were stable across multiple survival thresholds, we repeated the comparative network analysis in the two discovery datasets across multiple survival thresholds, which we ranged from zero to the maximum follow-up, with intervals of 0.1 years, keeping comparisons that had group sizes of at least ten patients. This figure reports the -log10FDR and GSEA enrichment scores (ES) for the different thresholds, for A) discovery dataset 1 and B) discovery dataset 2. Significant results (FDR<0.05) are highlighted with red circles.</p>