Ventricular-Subventricular Zone Contact by Glioblastoma is Not Associated with Molecular Signatures in Bulk Tumor Data

Whether patients with glioblastoma that contacts the ventricular-subventricular zone stem cell niche (VSVZ + GBM) have a distinct survival profile from VSVZ − GBM patients independent of other known predictors or molecular profiles is unclear. Using multivariate Cox analysis to adjust survival for widely-accepted predictors, hazard ratios (HRs) for overall (OS) and progression free (PFS) survival between VSVZ + GBM and VSVZ − GBM patients were calculated in 170 single-institution patients and 254 patients included in both The Cancer Genome (TCGA) and Imaging (TCIA) atlases. An adjusted, multivariable analysis revealed that VSVZ contact was independently associated with decreased survival in both datasets. TCGA molecular data analyses revealed that VSVZ contact by GBM was independent of mutational, DNA methylation, gene expression, and protein expression signatures in the bulk tumor. Therefore, while survival of GBM patients is independently stratified by VSVZ contact, with VSVZ + GBM patients displaying a poor prognosis, the VSVZ + GBMs do not possess a distinct molecular signature at the bulk sample level. Focused examination of the interplay between the VSVZ microenvironment and subsets of GBM cells proximal to this region is warranted.

Neural stem cell niches are hypothesized to support malignant gliomas. The two widely recognized neural stem cell niches are the ventricular-subventricular zone (VSVZ) 1 and the subgranular zone (SGZ) 2 . The former is located in the lateral linings of the lateral ventricles, and the latter is located in the dentate gyrus of the hippocampus 3,4 . While no correlation between patient outcome and glioblastoma (GBM) involvement with the SGZ is evident 5 , prior observations suggest that VSVZ contact by GBMs negatively impacts patient survival 5,6 .
These observations have rapidly led to at least 12 clinical studies assessing the benefit of incorporating VSVZ radiation in the standard GBM therapy regimen 7,8 . Eight of these studies did not show any benefit, and the benefit was minimal in the remaining 4. This lack of a therapeutic effect of VSVZ radiation is currently unexplained. Efforts to understand its basis must first address two gaps in our knowledge. First, a critical evaluation of the survival effect of VSVZ contact should ideally be adjusted for widely-recognized prognosticators of GBM patient survival, including extent of resection, tumor volume, and molecular features such as IDH mutation status, glioma CpG island methylator phenotype (G-CIMP), and MGMT promoter methylation status 5,6,9 . Second, whether GBMs with VSVZ contact or involvement (here termed VSVZ + GBMs) are molecularly different from VSVZ − GBMs is unknown. To date, there is little evidence to indicate whether VSVZ + GBMs are enriched for a specific molecular subclass or other genomic signature relative to VSVZ − GBMs [10][11][12][13][14][15][16][17] .
To address this critical gap in our understanding of the clinical and molecular differences in VSVZ + GBMs and VSVZ − GBMs, the association between patient survival and glioblastoma contact with the VSVZ was rigorously and comprehensively tested in two independent patient datasets. Further, computational analyses Scientific RepoRts | (2019) 9:1842 | https://doi.org/10.1038/s41598-018-37734-w were conducted on the molecular data available in the TCGA to identify any evident molecular signatures of VSVZ + GBMs and/or VSVZ − GBMs.

Methods patient Datasets and Clinical Data Collection.
Approval from an ethical standards committee (the Institutional Review Board) to conduct this study was received (Study IRB# 161891). Patient consent was waived. 170 consecutive adult (>18 years of age) patients who received a cytoreductive, maximal safe resection of a supratentorial GBM between 2011 and 2017 were identified (33% overlap with prior data 5 ). Following resection, all patients were considered for treatment with radiation and temozolomide according to the Stupp protocol 18 . Their clinical course was followed up to July 2017. Their age, preoperative Karnofsky performance status score (KPS), whether they received postoperative radiotherapy and temozolomide and completed the Stupp protocol regimen, their GBM molecular status (i.e., MGMT promoter methylation and IDH1/2 mutation), and overall (OS) and progression free (PFS) survivals were collected. Another 254 patients from The Cancer Imaging Archive (TCIA) 19 with contrasted brain imaging and corresponding OS data in The Cancer Genome Atlas (TCGA-GBM) database were identified (Table S1) 20 . Their demographic, clinical, and molecular information was obtained from the TCGA-GBM database 20 . Detailed treatment information on these patients was retrieved from Level 1 clinical data available from the Broad Institute's Genome Data Analysis Centers Firehose (http://firebrowse.org/). Patients who received adjuvant radiation of at least 60 Gy and completed at least 6 cycles of adjuvant temozolomide were noted. All data generated or analyzed during this study are included in this published article (and its Supplementary Information files) and are publicly available from the TCIA/TCGA databases. An analyzed institutional clinical dataset is available from the corresponding author on reasonable request.

Radiographic Data Collection.
Magnetic resonance images of the brain were available for all patients, except for 3 in the TCIA/TCGA-GBM dataset who had computed tomography images. The initial preoperative post-contrast brain imaging of all patients was assessed for VSVZ contact of GBM independently by two reviewers (neurosurgeon and a board-certified neuroradiologist) without knowledge of patient outcome. Using OsiriX Lite software (version 9.4, Pixmeo, Geneva, Switzerland), VSVZ + GBMs were identified by the contact or involvement of the post-contrast tumor enhancement with the lateral ventricular ependyma 5,6 , the location of the VSVZ in human brain 4 . Agreement was 86%. Disagreements were resolved by a consensus radiological review. Figure 1 depicts representative images of VSVZ + GBMs (Fig. 1A) and VSVZ − GBMs (Fig. 1B) from the TCGA/TCIA-GBM dataset. GBM contact with the ependyma of the third (n = 5 and n = 6 in the institutional and TCGA/TCIA-GBM datasets, respectively) or fourth ventricles (n = 2 and n = 0, respectively) was not considered when determining VSVZ contact status. Tumor volume was calculated using the semiautomated volume rendering function in OsiriX Lite after delineating the outer edge of a tumor's contrast enhancement 5 . In cases of multifocal tumors, their volumes were summed. For all the above assessments, axial imaging sequences were used; when clarification was needed, they were reformatted into coronal or sagittal sequences. In the institutional dataset, extent of resection was assessed independently by a neurosurgeon and neuroradiologist using postoperative, post-contrast magnetic resonance imaging obtained within 24 hours after the operation. Gross total (GTR) and subtotal resections (STR) were judged respectively by the absence or presence of residual tumor contrast enhancement. A portion of the STRs were deemed near-total resections (NTRs) if there was less than 5% residual tumor or if the neuroradiologist could not definitely exclude minuscule residual. The time at which the first postoperative, radiographic evidence of tumor recurrence or progression (PFS) agreed upon by both a neuroradiologist and neuro-oncologist (i.e., ruling out radiation necrosis or pseudoprogression) was also noted. PFS data were missing in 68 patients (26.8%) in the TCIA/TCGA-GBM dataset. Because two-thirds of these patients (23,  in the multivariate Cox analyses was tested by assessing the significance of the relationship between Schoenfeld residuals and time for the overall model. If this assumption was not met, a time-dependent covariate in the model was incorporated as a piecewise function with regards to time to confirm reproducibility of the results 25 . Survminer R package (Version 0.4.0) was used to perform these analyses, calculate median survival times, and plot right-censored Kaplan-Meier curves.
Molecular Data Analyses. Differences in gene mutations and copy number alterations between VSVZ + GBMs and VSVZ − GBMs were explored using Fisher's exact test. Differences in gene methylation, gene expression, and protein expression between VSVZ + GBMs and VSVZ − GBMs were assessed using the Mann-Whitney U test due to their global non-normal distributions. Further computational analyses conducted included: a weighted gene co-expression network analysis 26 ; partial least squares followed by logistic regression 27 to detect linear combinations of gene methylations, gene expressions, and protein expressions predictive of VSVZ + GBMs and VSVZ − GBMs; nonlinear dimensionality reduction of the multi-dimensional gene methylation, gene expression, and protein expression datasets using t-SNE 28 to segregate VSVZ + GBMs and VSVZ − GBMs in high-dimensional space; and unsupervised consensus clustering to reveal whether gene and protein expression clustered based on VSVZ contact 29 . Methods for these analyses are detailed in Supplemental Methods. Of the several gene expression datasets, Affymetrix HT Human Genome U133 was used to represent results. Results of these analyses on other datasets are available upon request.    VSVZ contact by GBM was also noted to be an independent predictor in the TCIA/TCGA-GBM dataset. Compared to the institutional dataset, this dataset included proportionally fewer patients who received temozolomide and radiation therapy and more patients with MGMT promoter methylated GBMs (Table 1) Fig. 3D). Because IDH mutation status was unavailable in 51 (20%) and 41 (19.7%) patients in the OS and PFS TCIA/TCGA-GBM datasets, respectively, its surrogate G-CIMP status (unavailable in 5 (2%) and 2 (1%) patients, respectively) was used to adjust the effect of VSVZ contact status.

VSVZ Contact Independently Stratifies
Confounder-based subgroup survival analyses revealed that VSVZ + GBMs were associated with decreased overall and progression free survival in the majority of the subgroups (Fig. 4). Interestingly, in both datasets, the effect of VSVZ contact on survival was greater in younger patients 50 or less years of age. This effect was maintained in an adjusted survival analysis. Patients   (Table S4), copy number variation (Table S5), gene expression (Fig. 5A, Table S6), protein expression (Table S7), or DNA segment methylation (Table S8) met the statistical threshold to be confidently deemed as different between VSVZ + GBMs and VSVZ − GBMs groups with the sample size used. Re-analysis pre-filtering 50% of the genes with minimal global variation, which decreases the penalty imposed by multiple hypothesis adjustment to increase discoveries 30 , did not alter these results. Therefore, alternative computational approaches were considered. A weighted gene co-expression network analysis 26 identified 14 modules of co-expressed/coregulated genes (Fig. 5B) but none were enriched in either VSVZ contacting or non-contacting GBMs (Fig. 5C,D, Table S9). A partial least squares followed by logistic regression modeling 27 using linear combinations of gene and protein expression found no combinations to be predictive of VSVZ + GBM and VSVZ − GBM status (Fig. 6A,B). In addition to these above linear analyses, nonlinear dimensionality reductions were performed using t-SNE 28 , but VSVZ + GBMs remained indistinguishable from VSVZ − GBMs in high-dimensional space based on their gene or protein expression or DNA methylation (Fig. 6D,F). Lastly, unsupervised consensus clustering was performed to identify potential distinct GBM clusters. Although GBMs could be successfully clustered based on gene or protein expression, no cluster was enriched in VSVZ + GBMs or VSVZ − GBMs (Fig. 7). Computational molecular analyses restricted to only G-CIMP negative and IDH wild-type patients were no different from those seen in the entire dataset (Tables S11-S17).
To confirm our approaches, we tested the results of computational analyses such as gene co-expression and consensus clustering on gene expression datasets and found significant correlations with known canonical transcriptional GBM subtype classifications (Fig. 5E,F, Tables S9 and S10). None of these subtypes were predominant in either VSVZ + GBMs or VSVZ − GBMs (Χ 2 P = 0.11).

Discussion
Our survival analysis of VSVZ + GBMs and VSVZ − GBMs was adjusted for universally-accepted predictors of survival in GBM patients, including patient performance status (i.e., KPS), tumor volume, and extent of resection, which are potential confounders of an analysis focused on this region. Lower KPS has been observed in patients with VSVZ + GBMs 5,9,11,16 , likely due to their more central location and larger size which could impair  In addition, their larger volumes and deeper location may also decrease the likelihood of gross total resections 9,16,31-33 . Unlike prior efforts 5,6,15,16 , our analysis accounted for these and other molecular predictors and demonstrated that contact with the VSVZ by GBM is an independent, negative prognosticator. These results were consistent between our institutional and the TCIA/TCGA-GBM datasets.
The distinct survival phenotypes of VSVZ + GBMs and VSVZ − GBMs and the absence of molecular signatures which identify them offer an important guide to future research aimed at understanding the role of VSVZ contact in GBM pathobiology. These results suggest that VSVZ + GBMs and VSVZ − GBMs are not transcriptionally or genomically distinct when considered as bulk samples; therefore, the unique cytoarchitecture and molecular properties of the VSVZ deserve increased attention. Additionally, enrichment for a region-specific phenotype may not be detectable in the bulk tumor samples examined here.
The growth factor-rich environment of the niche may uniquely support and/or enhance the neoplastic potential of GBMs. The normal VSVZ supports cells which are responsive to epidermal growth factor (EGF), platelet-derived growth factor (PDGF), and Ephrins, among other growth factors and chemokines 34,35 . With the exception of EGFR amplification, a greater proportion of receptor tyrosine kinase gene amplifications, specifically PDGFR-α, EPHB3, and KIT, was noted in VSVZ + GBMs, but did not meet our adjusted statistical thresholds (Tables S5 and S12). These results suggest that VSVZ-enriched growth factors could support the aberrant, neoplastic growth signaling of GBM cells through these amplified growth factor receptors. GBM cells are similarly known to be attracted to the niche by VSVZ growth factors. This has been observed radiographically in humans 5,36 and in mouse xenograft models, which revealed specific mediators including VSVZ derived stromal cell-derived factor 1 (SDF1 or CXCL12) 37 and neurite growth-promoting factor 1 (NEGF1 or Pleiotrophin) 38 .
Interestingly in a mouse model, SDF1 released by the VSVZ mediated resistance to radiation therapy in the GBM cells located in the VSVZ 39 . This may partly explain why 8 of the at least 12 clinical studies conducted to date failed to demonstrate a benefit of incorporating VSVZ radiation in the standard GBM therapy regimen 7,8 . Upon the arrival of cancer cells in this niche, the VSVZ's fertile microenvironment, cytoarchitecture (including a well-characterized gap layer in the adult human brain), and extensive contact with the circulating CSF likely make it uniquely permissive for cell migration, including sub-ependymal spread [40][41][42] and widespread dissemination. Several observations support the hypothesis that the VSVZ permits spread of glioma cells. First, VSVZ contact is associated with multifocal GBM 5,43-45 . Second, VSVZ + GBMs are associated with distant recurrences post-therapy 43,44,46,47 . The exact VSVZ structural components that allow GBM cells to spread and disseminate once in this region remain unknown. Although multiple factors controlling the rostral migration of young neurons in this niche have been identified in the mouse, it is less clear which of these factors are persistently expressed in adult human brain. Finally, surgical violation of the VSVZ resulting in an entry of tumor cells to the circulating CSF is also associated with widespread dissemination and distant recurrences 31,48,49 .
VSVZ neurogenesis and its other reparative functions decline precipitously with age in humans 50,51 . The niche may therefore be more fertile in younger patients, meaning that GBMs may draw more malignant potential from components of the VSVZ niche, such as the cerebrospinal fluid, in these cases. The age-dependent effect of VSVZ contact by GBMs on patient survival, where younger VSVZ + GBM patients in both datasets had a significantly greater rate of mortality compared to older VSVZ + GBM patients, supports this inference. However, a more detailed examination of the effects of host age will likely be necessary to investigate potential molecular mechanisms driving this clinical finding.
This study reveals contact with the V-SVZ as an independent predictor of outcome without obvious molecular correlates, indicating that VSVZ + GBMs and VSVZ − GBMs may not be intrinsically different on the analysis platforms tested. One hypothesis derived from these findings is that the VSVZ, and the secreted factors enriched within it, may elicit stem-like functional features in otherwise similar tumor cells. Alternatively, the VSVZ may support transcriptionally distinct subsets of GBM cells which are not readily detectable in bulk analyses. Therefore, further aggregate or single-cell analyses coupled with position-specific sampling are warranted. In support of this latter hypothesis, we note that a focused study of sphere cultures derived from subregions of GBM tumors suggested possible enrichment for the mesenchymal subtype in cultures derived from VSVZ-proximal Figure 7. Clustering demonstrates no correlation with ventricular-subventricular zone-contacting glioblastomas (VSVZ + GBMs) or VSVZ − GBMs in the TCIA/TCGA samples. Gene (A) and protein (B) expression datasets clustered into two and four consensus clusters (CC), respectively, are depicted. The relationship between samples and clusters is depicted with a dendrogram above the image. Underneath, the three bars are used to represent the classification of each sample using a color. Subtype refers to the transcriptional classification. Preponderance of neither VSVZ + GBMs nor VSVZ − GBMs is noted in the two gene or four protein expression clusters.
Scientific RepoRts | (2019) 9:1842 | https://doi.org/10.1038/s41598-018-37734-w regions in comparison to VSVZ-distant regions of the same tumor 52 . However, additional research on this topic is needed to separate the potential contribution of any resident non-cancerous neural stem cells, which share many features with the mesenchymal signature, to such molecular analyses. Future studies examining the cytoarchitectural structure and growth factor-rich environment of the VSVZ together with its influence on the functional phenotypes of GBM cells (e.g. quiescence, radiation resistance, and metabolic alterations) 53 are warranted to shed light on the distinct survival profiles of VSVZ + GBMs and VSVZ − GBMs. Finally, single-cell rather than homogenizing methods will be of value when identifying the exact GBM cells influenced by the VSVZ. In addition to single-cell RNA sequencing or lineage barcoding, capturing intracellular signaling and the dynamic modulation of stem-like features by VSVZ growth factors in GBM cells will likely generate valuable information. These investigations will be key components in the search to identify the mechanistic underpinnings of the observed survival phenotype, dissemination, and therapy resistance.

Conclusion
This comprehensive survival and molecular analysis finds that survival of GBM patients is independently stratified by VSVZ contact of the tumor. Specifically, patients with VSVZ + GBMs have a significantly lower survival. A thorough analysis of the TCGA molecular data did not reveal a molecular signature specific to the VSVZ + GBMs.