Despite recent advances in multimodal treatment, de novo glioblastoma (GBM) World Health Organization (WHO) grade IV remains one of the most intractable human cancers, with a median survival of less than 15 months [1] and few long-time survivors [2]. Extensive efforts have been made to maximize the extent of resection (EOR) with simultaneous preservation of neurologic function and quality of life [3, 4]. At the same time, there are apparent advances in postoperative radiotherapy, chemotherapy, antiangiogenic therapy, immunotherapy, and targeted therapies [5, 6]. Nevertheless, recurrence occurs almost inevitably, in most cases adjacent to the resection cavity, leading to non-standardized salvage therapies and ultimately to death. Treatment failure has been attributed in part to the fact that GBM is not a monoclonal disease but is characterized by intra- and intertumoral heterogeneity, resulting in divergent clinical presentation and response to treatment. In this context, several molecular subtypes have been identified [7, 8] with distinct driver mutations, prognostic impact, and prediction of treatment response, including a glioma-CpG island methylator phenotype (G-CIMP) [9] which is highly dependent on the presence of mutations in the isocitrate dehydrogenase 1 (IDH1) gene and is associated with improved patient outcome [10]. Tumor location with respect to distinct brain regions reflects another important aspect of intertumoral heterogeneity. In particular, vicinity of de novo GBM to the subventricular zone (SVZ) lining the lateral ventricles, one of the persisting neurogenic regions in the adult human brain [11, 12], has been linked to inferior patient outcome [13] and a distinct growth pattern. Lim et al. initially reported a series of 53 de novo GBMs that were preoperatively classified by their vicinity to the SVZ into four groups. Group I consisted of GBMs with the contrast-enhancing (CE) lesion contacting the SVZ and infiltrating the cortex, group II of tumors contacting the SVZ but not involving the cortex, group III of GBMs not contacting the SVZ but involving the cortex, and group IV of tumors neither contacting the SVZ nor infiltrating the cortex [14]. Group I GBMs were most likely to be multifocal at first diagnosis and to recur distant from the resection cavity, while group IV GBMs were always solitary lesions with recurrences exclusively adjacent to the primary site. The authors concluded that GBMs with (SVZ+) and without (SVZ–) contact to the SVZ might arise from different cells of origin and that SVZ+ GBMs might reflect tumors with a high content of SVZ stem cells that have undergone malignant transformation, a hypothesis that has rarely been explored in detail. In one of the few studies addressing this issue so far, Kappadakunnel et al. failed to identify a stem cell-derived gene signature by means of a DNA microarray analysis of 47 GBMs classified according to their relationship to the SVZ [15]. In contrast, in a phylogenetic approach making use of intraoperative fluorescence-guided multiple sampling (FGMS) of human GBMs and their adjacent (fluorescent) SVZs, Piccirillo et al. were able to identify the SVZ as a reservoir of malignant precursor clones in the majority of tumors analyzed [16]. Accordingly, several clinical studies have provided evidence that targeting the ipsilateral SVZ by irradiation is associated with superior survival in patients with GBM [17, 18], especially in combination with gross total resection (GTR) [19], supporting the hypothesis that the SVZ plays a role in GBM formation and propagation.

Even though SVZ+ GBMs appear to be associated with a distinct clinical and radiographic behavior, little is known about the molecular phenotype underlying these characteristics and potential biomarkers linked to this particular tumor location. Therefore, the aim of this study was to identify a gene signature distinctive of de novo GBM in vicinity to the SVZ and to discover location-dependent genes with a potential prognostic impact. Noteworthy, validation of differential gene expression and prognostic relevance was performed in a confirmatory patient cohort restricted to IDH wild-type (wt) GBM, excluding the unique molecular and prognostic phenotype of IDH mutant (mt) GBM.

Intertumoral heterogeneity is one of the mainstays of treatment failure in GMB; thus, there is a need for individualized prognostication and treatment approaches. Tumor location is one important aspect that clearly determines treatment options, functional outcome, and quality of life. However, it is largely unknown whether tumor location is linked to a distinct molecular phenotype. In this study we sought to compare transcriptomes of GBMs with and without spatial relationship to the SVZ in order to identify location-dependent gene signatures and prognostic markers. In contrast to previous studies, location-dependent candidate genes identified in a discovery set were validated in an independent patient cohort comprising exclusively IDH wt GBM. Thereby, we sought to exclude the unique molecular and prognostic phenotype of IDH mt GBM [9].

Hierarchical clustering of microarray data revealed two main transcriptional profiles that perfectly matched the allocation of tumors according to their vicinity to the SVZ in general and to group II (GBMs contacting the SVZ only) and group III (GBMs contacting the cortex only) tumors in particular. MetaCore™ enrichment analysis linked these SVZ-dependent transcriptomes to major pathways involved in cell growth and motility, angiogenesis, immune modulation, and Notch signaling, one of the major developmental pathways involved in neural stem cell (NSC) maintenance and gliomagenesis [26, 27]. Importantly, no overlap was found between our location-specific transcriptional profiles and the four molecular subgroups described by the TCGA [8]. In face of our study’s relatively small case number, tumor vicinity to the SVZ does not appear as a determining factor of the TCGA’s molecular phenotypes.

Ever since the early reports that vicinity of GBM to the SVZ is linked to a distinct growth behavior and inferior patient outcome [13–15], it has been hypothesized that SVZ+ GBM may arise from transformed NSCs originally residing within the SVZ [11, 12]. Indeed there is evidence from rodent studies that inactivation of typical tumor suppressor genes (TP53, NF1, PTEN) allocates formation and early progression of high-grade astrocytoma to neural stem/progenitor cells within the SVZ [28–30]. In humans, intraoperative fluorescence-guided multiple sampling (FGMS) along a spatial gradient within the tumor mass and the adjacent (fluorescent) SVZ recently allowed for a phylogenetic reconstruction with SVZ-derived tumor precursor cells giving rise to the tumor mass in the majority of GBM patients analyzed, and thus, for the first time, substantiated a role of the SVZ in gliomagenesis in humans [16]. These observations also suggest that SVZ+ GBMs are enriched in NSCs and BTICs, a hypothesis that has not yet been explored in much detail. In a microarray analysis of 47 GBMs, Kappadakunnel et al. did not find a correlation between 7 selected stem cell-related genes of interest (PROM1, MELK, BMP4, ETF2, MAPK8, OLIG2, NES) and tumor location [15]. Instead, 7 of the 33 genes overexpressed in group II tumors were related to immune signaling (FCGR3A, HLA-DRB5, BCL6, FCGR3B, MAFB, HLA-DRA, HLA-E), a finding which was underscored by our MetaCore™ pathway analysis.

To further investigate a potential stem cell origin of SVZ+ GMB, we searched our microarray data for location-specific differential expression of a panel of 47 well-known NSC, BTIC, and lineage markers as well as key players of three development pathways (Notch, Wnt, and Sonic Hedgehog). Since group I GBMs that are defined as SVZ+ GBMs consist of voluminous tumors that reach from the cortical surface to the SVZ and, in theory, may reflect tumors that originate from the cortex rather than the SVZ, we decided to compare group II and group III tumors only, since this reflects the most concise regional separation between GBMs with (group II) and without (group III) SVZ contact. Differential regulation of gene expression was observed for a subset of genes only (FZD6, PTCH2, NEFH, NOTCH2, HES4, DLL3), including three genes (NOTCH2, HES4, DLL3) involved in Notch signaling. Remarkably, location-dependent differential gene expression was detected at all key hierarchies of the Notch pathway (depicted in Fig. 4). It is well known that Notch signaling drives NSC maintenance and differentiation of neural progenitor cells into astroglia (reviewed in [27]). Moreover, aberrant Notch pathway activation contributes to formation and propagation of primary GBM [31], possibly through propagation of the BTIC pool [32], and therapeutic inhibition of Notch signaling is under investigation both in vitro and in vivo (reviewed in [33]). In this regard, our study provides the first link between Notch expression and tumor location. With the Notch ligand DLL3 and the transcription factor HES4 we found pivotal upregulated genes that later proved to be prognostic in a multivariate setting. However, our data are restricted to the transcriptional level, and functional analyses are needed to shed further light on location-specific Notch pathway activation. Nevertheless, neither this analysis nor that of Kappadakunnel et al. found unequivocal evidence of a (cancer) stem cell signature in SVZ+ GBM [15]. This may account for the fact that differential expression was assessed on a transcriptional level only and that in-depth analysis of a larger patient sample, in particular combined with investigation of protein expression, may yield unambiguous results. A sampling error may also come into play. Since all samples were retrospectively identified from our tumor bank, it is impossible to reconstruct the exact position of sampling, in particular with regard to distance from the SVZ. Particularly in group I GBM, the most voluminous location-specific subgroup extending from the SVZ throughout the white matter to cortical areas, there may well be a spatial gradient in the enrichment of tumor tissue with NSCs and BTICs, a factor impossible to take into account unless tissue sampling is done in a prospective manner with multiple samples derived from the same tumor and the regions of interest defined on perioperative imaging [16, 34]. Ultimately, it would be worthwhile to compare transcriptomes from SVZ+ and SVZ– GBMs to those derived from non-malignant human SVZ to elucidate the role of the SVZ in human gliomagenesis. However, to the best of our knowledge, array data from human SVZ tissue have been lacking until now.

To attenuate the sampling issue, we ensured that all GBM tissues analyzed were homogeneous throughout the location groups in terms of tumor content and interplay from the microenvironment (Additional file 1: Figure S1A–D) and validated HES4 and DLL3 expression together with 11 other top differentially expressed candidate genes derived from microarray analysis in an independent set of 142 IDH1 wild-type GBMs (validation cohort). Hence, we were able to confirm SVZ-dependent expression of HES4 and DLL3. HES4 was significantly overexpressed in both SVZ+ GBMs and group II GBMs. HES4 is one of seven HES target genes of Notch signaling that serve as tissue-specific transcription repressors upon Notch pathway activation, leading to inhibition of cell differentiation and maintenance of stem cell features [35]. Little is known about the particular function of HES4 in stem cells and cancer. HES4 was shown to keep retinal precursor cells of the Xenopus ciliary margin zone in an undifferentiated and slowly proliferative state [36]. In human B cells, HES4 inhibits early differentiation and acts as a tumor suppressor with epigenetic silencing in B cell acute lymphoblastic leukemia [37]. Recently, HES4 has been established as a biomarker in advanced solid tumors, predicting treatment response to Notch pathway inhibition by gamma-secretase inhibitors [38, 39]. Notably, our analysis constitutes the first report on HES4 interaction in human brain tumors.

DLL3 is the second Notch pathway key player for which a strong trend towards overexpression in group II GBMs was confirmed in our analysis. DLL3 is a direct ligand to Notch receptors, with conflicting data about its activating or rather inhibiting role in Notch signaling [40–44]. Likewise, DLL3 has been described in the context of proneural GBM with inconsistent expression patterns: according to Phillips et al. [7] and Verhaak et al. [8], overexpression of DLL3 is a hallmark of proneural GBM, while Cooper et al. [45] report loss of DLL3 in proneural GBM. In the meantime, identification of an IDH-dependent G-CIMP phenotype has further separated the initially favorable appraised proneural molecular subgroup into GBMs with an IDH-mutant, G-CIMP-positive phenotype exhibiting a highly favorable prognosis and an IDH-wild-type, non-G-CIMP phenotype with an exceedingly dismal prognosis [9]. This is of importance, since none of the aforementioned studies stratified expression and survival data for IDH mutation status, while our study comprised IDH1 wt GBM only. Indeed, DLL3 was significantly overexpressed in the 10 tumor samples of our microarray cohort assigned to the proneural subtype (p = 0.0078; Mann-Whitney test; data not shown), but no difference in OS was observed between proneural and other subtypes (p = 0.4776; log-rank test; data not shown).

We also observed a strong trend towards upregulation of NTRK2 and downregulation of PIR in group II GBMs (contacting the SVZ only). NTRK2 encodes for the neurotrophic tyrosine kinase, receptor, type 2, also known as Trk-B. Neurotrophins and their receptors are crucial for cell growth, survival, and apoptosis in the nervous system [46], but expression has been observed in glioma as well [47], even though their functional role is largely unknown. Activation of Trk-B and Trk-C has been shown to promote growth and survival of BTICs independent of epidermal growth factor (EGF) and basic fibroblast growth factor (bFGF) [48]. In our study, NTRK2 overexpression was found in GBMs with SVZ contact, possibly maintaining BTIC growth as well. PIR is another location-specific candidate gene with downregulation observed in group II GBMs. It encodes for the iron-binding nuclear protein pirin, a transcriptional regulator, and has been described as an oncogene [49] and promoter of metastatic tumor growth [50] on one hand and as a tumor suppressor gene on the other [51] in many solid cancers, but never before in glioma. In acute myeloid leukemia (AML), PIR was linked to terminal differentiation of myeloid precursors with a downregulation of PIR possibly related to the differentiation arrest observed in AML [52]. By contrast, overexpression of PIR was involved in inhibition of cellular senescence in melanocytic cells, resulting in transformation to melanoma [53]. In GBM, the functional role of PIR has yet to be discovered.

Besides their SVZ-dependent expression, HES4, DLL3, PIR, and NTRK2 also conferred a significant impact on patient survival, together with known clinico-pathological prognosticators. Importantly, the reported adverse effect of SVZ involvement on OS [13] was confirmed in our study sample by multivariate analyses. In line with its distinct expression in GBMs contacting the SVZ, HES4 turned out to be the most robust prognostic marker with overexpression related to adverse OS, even overriding the prognostic effect of all other molecular markers in the multivariate analysis. As elucidated above, HES4 is a novel molecular marker in GBM and underscores the biological and clinical role of Notch pathway activation in primary GBM, in particular in tumors involving the SVZ. Likewise, overexpression of DLL3, the second molecular marker involved in Notch signaling, resulted in significantly reduced OS and PFS in the univariate analysis and turned out to be an independent prognostic marker of inferior PFS in the multivariate analysis. Moreover, high expression of PIR and NTRK2 was associated with superior OS. It is noteworthy that in patients with GBMs contacting the SVZ (SVZ+ GBMs, group II GBMs), NTRK2 expression levels were able to predict OS. The beneficial effect of NTRK2 overexpression on OS confirms recent data reporting that loss of mRNA expression of both NTRK1 and NTRK2 correlates with poor prognosis in patients with high-grade glioma [54], but our analysis adds a location-specific link to the picture. To our knowledge, this is the first report on PIR as a prognostic marker in GBM, with evidence of SVZ-dependent differential expression, and it is worth further functional analysis to investigate its role as an oncogenic or tumor repressive factor. In addition to these four genes with SVZ-dependent differential expression and concurrent prognostic impact, we also identified IGFBP5 expression to be inversely related to OS, even though a location-specific expression could not be established. There is sparse evidence from the literature that overexpression of IGFBP5 increases in a WHO grade-dependent fashion with highest expression observed in GBM [55, 56] and a strong trend linking overexpression to adverse OS [55]. The functional role of IGFBP5 has not been fully clarified, but it may play a role in tumor dormancy, among others in GBM [57].

In summary, this study revealed inherent transcriptional differences of GBMs, depending on their vicinity to the SVZ. Interestingly, all four genes with simultaneous SVZ-dependent differential expression and significant prognostic impact were characterized by their involvement in stem cell maintenance. Two of these genes (HES4, PIR) have never before been reported in the context of gliomagenesis and deserve further functional exploration. Importantly, Notch signaling was an outstanding feature of SVZ+ GBM, with the two key players HES4 and DLL3 identified as location-specific prognosticators. Further work will be required, but this finding suggests that SVZ+ GBM might profit most from the therapeutic Notch inhibition that is currently under investigation in clinical trials. Mounting evidence is in favor of location-tailored therapies, since irradiation of the ipsilateral SVZ as a potential BTIC niche has been shown to have a positive impact on patient outcome [17, 18], particularly in patients with GTR [19]. SVZ-specific targeted molecular therapies might add another important piece to the picture.