We first demonstrate that in addition to including multiple

We first demonstrate that, in addition to including multiple layers of data, aSICS outperforms a standard method in terms of consistency between two glioma data sets. Next, we use aSICS to build a multi-layered model of regulation of glioma subtypes. Our model identifies both known and novel regulators of mesenchymal, proneural and classical subtypes. Among the predicted regulators, Annexin A2 (ANXA2) stands out as an epigenetically controlled master regulator of mesenchymal transformation in glioma, associated with patient survival. To validate the functional relevance of this model prediction, we abrogate ANXA2 expression in patient-derived, IDH1 wildtype glioma cancer stem cells, leading to loss of mesenchymal signature genes. Interestingly, ANXA2 loss led to reduced phosphorylation of the previously described key regulator of mesenchymal transformation, STAT3 (Carro et al., 2010). In-depth analysis of two independent patient cohorts further supports that ANXA2 suppression in lower grade glioma is likely explained by methylation induced by IDH1 mutation. However, ANXA2 also retains IDH1-independent prognostic power in IDH1 wildtype higher grade glioma. Analysis of heterogeneous glioma materials, both surgical samples from different ganciclovir regions and single cell data, is used to show that ANXA2 correlates regionally with mesenchymal transformation.
Methods
Results
Discussion Our MRI-localized regional analysis of ANXA2 expression in GBM showed a variation across the tumor, with elevated expression in regions expressing a mesenchymal markers. This finding, in turn, was consistent with a re-analysis of the data by Patel et al, in which ANXA2 expression predicts mesenchymal transformation in individual cells (c.f. Supplementary Fig. 4). Reduced ANXA2 expression in the edema zone compared to contrast-enhancing zone and core of mesenchymal tumors might be related to the particularly heterogeneous nature of this region in which ANXA2-expressing neoplastic cells are intermixed with non-neoplastic cells without ANXA2 expression (Gill et al., 2014). GBM tumors often contain cells with distinct transcriptional signatures; such heterogeneity might reflect the evolution process of tumor cells by the accumulation of temporal mutation events (Ozawa et al., 2014; Patel et al., 2014; Sottoriva et al., 2013). The different level of ANXA2 expression in the tumor may thus suggest that it is an interesting target to re-program mesenchymal subsets of GBM cells. Functional experiments in BTSCs showed that ANXA2 is a causative regulator of mesenchymal transformation, which both regulates and co-depends on previously described regulators. Firstly, ANXA2 knockdown and overexpression in BTSCs suppressed and induced mesenchymal genes, respectively. In BTSC168 ANXA2 knockdown resulted in a more pronounced loss of the mesenchymal signature by Phillips and Verhaak compared to the other tested cell line, BTSC161 (Fig. 4c,d, Supplementary Fig. 5c); this could reflect different gene expression properties of the two cell lines. Interestingly, BTSC168 were established from a recurrent GBM, so they might display more mesenchymal features consistent with previous indication that glioblastoma tend to shift to a mesenchymal phenotype upon recurrence (Phillips et al., 2006). Exogenous expression of ANXA2 appears to induce a positive feedback loop that induces endogenous ANXA2 in BTSCs. This is interesting in the light of ANXA2 presence in microvesicles secreted by GBM cells which are able to deliver material into neighboring cells (Bronisz et al., 2014). Potentially, such a positive feedback loop might represent a mechanism of paracrine interaction between tumor cells to coordinate cellular behavior across the cell population. Effects of ANXA2 on cellular properties in GBM has been previously reported (Onishi et al., 2015; Tatenhorst et al., 2006; Zhai et al., 2011), although these studies only used human GBM cell lines or murine models. Here, we investigated the ANXA2-dependent mesenchymal signature modulation effect on cell behavior in BTSCs. As previously described for GBM cell lines (Tatenhorst et al., 2006; Zhai et al., 2011), ANXA2 expression levels also determine cellular proliferation and invasiveness in BTSCs. In BTSCs, loss of the mesenchymal signature prevented the cells from differentiating into mesenchymal progeny. Thus perturbation of ANXA2 translates into an actual change in cell phenotype by affecting key properties such as proliferation, invasion, and notably, cell fate.