The progression of a GBM depends on invasive growth

The progression of a GBM depends on invasive growth marked by mesenchymal cell features. It therefore remains a critical priority to expand our knowledge of how the mesenchymal features are modulated. We have employed aSICS as an integrative modeling method to identify regulators of cancer subtypes. Applied to GBM, our model implied that a number of methylation events, independently of IDH1 mutation, contribute to a mesenchymal/proneural axis in GBM. We confirmed ANXA2 as a key regulator of mesenchymal transformation and demonstrated its importance for viability, invasiveness and maintaining a mesenchymal gene signature. The results thus show that integrative modeling can uncover a new mesenchymal modulator. Additional predictions provide a rich source for future investigation (Table 1). Our results warrants further investigation of aSICS as a general tool to uncover cancer subtypes. Compared to existing approaches, aSICS includes broader spectrum of data types, and benchmarking confirmed reproducible performance between independent cohorts. Explorative analyses of three additional cancers reveal that TF, miRNA, mutations and DNA methylation regulators are also detected in breast, ovarian and colorectal cancers, suggesting that new regulators can be predicted (Supplementary Fig. 9). A second important venue of investigation will be to explore the impact of alternate classification schemes, or using the model to improve subclassifications. For instance, the absence of regulators of the neural subtype is notable and may indicate a lack of mechanistic support for this subtype. Furthermore, there are currently inconsistent observations regarding the prognostic value of mesenchymal signatures from individual biopsies (Noushmehr et al., 2010; Ozawa et al., 2014; Phillips et al., 2006; Verhaak et al., 2010). These observations motivate extended method development and broader applications of the aSICS framework, also taking into account tumor heterogeneity, reserved for future work. Furthermore, the bootstrapping framework can be generalized, e.g. to compute confidence intervals of connectivity scores, reserved for future work. The software itself is currently available in Matlab and can be obtained from the authors upon request. In addition to the finding observation that ANXA2 is a key regulator of mesenchymal targets, our analysis adds to the characterization of ANXA2 as a possible biomarker in solid tumors (Liu et al., 2015; Zhang et al., 2012). Based on our results, ANXA2 promoter methylation shows promise as a prognostic marker, and we also find that ANXA2 PR619 levels to be predictive of patient survival. When extending the analysis to lower grade glioma (LGG) and secondary GBM, which tend to be IDH1 mutant and hypermethylated, we noted elevated ANXA2 promoter methylation and suppressed expression. In the G-CIMP signature described by Noushmehr et al. (2010), ANXA2 is a methylated gene. Together, these observations imply that ANXA2 can be suppressed by IDH1 mutation in LGG but also that IDH1-independent mechanisms can modulate ANXA2 via methylation of its promoter in GBM cells. Although we did not observe any significant effect of ANXA2 on global DNA methylation, several G-CIMP target genes were among the genes affected by ANXA2 knockdown. Thus, it is possible that ANXA2 knockdown leads to a general shift in gene expression that effectively mimics the expression modulation normally achieved through DNA methylation in G-CIMP tumors. Since G-CIMP tumors are associated with the most favorable patients\' prognosis (Noushmehr et al., 2010), we speculate that the acquisition of a G-CIMP-like gene expression signature upon ANXA2 knockdown in BTSCs might explain the connection between ANXA2 methylation and survival in our two cohorts. IDH1 mutation correlates significantly with both ANXA2 methylation or expression. However, as estimated by a correlation analysis on TCGA cases, IDH1 status only explains circa 62% of the variation of ANXA2 promoter (cg08081036) methylation and 26% of the expression variation. Thus, removing IDH1 mutant cases from the analysis, a significant degree of correlation between ANXA2 expression and methylation is retained (r=−0.21, p=0.03). We further find that after removal of IDH1 mutant cases, ANXA2 expression is predictive of longer survival (Supplementary Fig. 10A). In addition, in the Freiburg patient survival analysis (Fig. 4e) ANXA2 methylation was not always associated with IDH1 status. In fact, in the group of secondary GBM (shown in Fig. 3b and in the survival analysis (Fig. 3e, indicted by a “x”) two patients with ANXA2 methylation but wildtype IDH1 status showed nonetheless a more favorable prognosis. Combination of ANXA2 expression with recently described prognostic indicators 1p19q co-deletion and TERT promoter mutation (Eckel-Passow et al., 2015) also suggested that ANXA2 expression adds independent prognostic power (Supplementary Fig. 10B), motivating combined analysis of ANXA2 and other indicators in larger cohorts. Together, our results therefore imply that while IDH1 mutation is a key mechanism behind ANXA2 suppression, particularly in LGG and secondary GBM, it is likely regulated by additional factors, and may have clinical promise as a prognostic complementary to IDH1 mutation, or as an indirect marker of G-CIMP cases.