Introduction
Glioblastoma (GBM) is the most common primary brain tumor and is among the most vascularized solid tumors [
1‐
4]. The dismal prognosis of patients with GBM warrants development of more effective therapies. Despite the general consensus that GBM comprises a tumor with extensive heterogeneity, all patients are currently still treated uniformly [
2,
5,
6].
Molecular subclasses have been described over the past decade based on gene expression patterns, but the discussion on the exact number of subclasses is still ongoing. Initially a set of three subclasses was identified [
7], but the definition was strongly refined, expanded to four subclasses and specific molecular aberrations were coupled to the subclasses by The Cancer Genome Atlas (TCGA) Network [
8]. The discussion is still not definite and only recently another proposition for subgroups of gliomas has been postulated [
9]. From the earlier profiling endeavors, however, three molecular subclasses can be distilled consistently which include the proneural (PN), classical (CLAS) and mesenchymal (MES) subclasses.
Thus far only few studies have reported on the angiogenic properties of the molecular subclasses. Platelet/endothelial cell adhesion molecule 1 (PECAM1), vascular endothelial growth factor A (VEGFA) and vascular endothelial growth factor receptor 1 and 2 (VEGFR1, 2) were reported to be upregulated in MES GBMs [
7]. Other results only have described differential responses of GBMs to anti-angiogenic treatment in both the pre-clinical and clinical setting. One study has addressed whether the differences in response were associated with the molecular subclasses and reported selected benefit for PN subclass [
10]. In an aortic ring assay anti-VEGFA treatment abrogated microvessel sprouting in one cell line but not in the other [
11], and tumor recurrences following anti-VEGFA treatment could be divided in distinct resistance phenotypes [
12].
Since the subclasses are characterized by molecular aberrations, differential employment of angiogenic signaling pathways could also be suggested based on the defining features of the subclasses. It has for example been shown that IDH1
mutant protein, which is exclusively expressed in PN GBMs, stabilizes HIF-1α-expression [
13]. HIF-1α on its turn is known as a potent inducer of angiogenesis via multiple signaling pathways [
14]. The expression of EGFRvIII, the constitutively active (mutant) form of
EGFR characteristic of CLAS GBMs, was on the other hand reported to promote angiogenesis by activating the IL-8 pathway and inducing expression of cytokines and interleukins [
15,
16].
To determine whether specific angiogenesis-related factors could serve as possible new candidates for GBM subclass-differentiated anti-angiogenic therapy, we assessed the vascular status of the subclasses using morphometry, immunohistochemistry (IHC) and polymerase chain reaction (PCR).
Methods
Patient population
Tissue samples from 30 patients diagnosed with GBM were selected for the Groningen cohort. All patients underwent neurosurgical debulking at the University Medical Center Groningen (UMCG) in the Netherlands in the period August 2006 to May 2012. The mean age at diagnosis was 55 and the higher incidence rate in males was reflected in the male to female ratio (67 % male) [
17]. The samples selected for this study, 10 per subclass, were all previously identified as PN (
IDH1
mut
), CLAS-like and MES-like subclasses [
18]. Due to the fact that this cohort of GBMs was not subclassified transcriptionally but instead through a protein-based approach, we refer to the subclasses of these tumors as PN
IDH1
mut
, CLAS-like and MES-like where the analyses concern these tumors. Archival tissue of all patients was handled according to the Dutch Code of Conduct for proper secondary use of human tissue (
http://www.federa.org).
Level 3 gene expression data was obtained of 146 GBM patients that were previously transcriptionally subclassified (core TCGA samples,
http://cancergenome.nih.gov/) [
8]. The age at diagnosis in this cohort was 55 and the male percentage was 62. Clinical characteristics of both cohorts are summarized in Table
1.
Table 1
Summary of characteristics of patients with GBM in the Groningen and Verhaak cohort
Number of patients | 30 | 146 |
Mean age at diagnosis (95 % CI) | 55 (49–60) | 55 (53–58) |
Median overall survival in days (range) | 526 (62–1447) | 361 (0–3524) |
Male sex (%) | 20 (67) | 91 (62) |
Female sex (%) | 10 (33) | 55 (38) |
Necrosis measurement on MRI scans
Pre-operative imaging of the Groningen cohort was used to measure the volume of central necrosis in relation to total tumor volume. Volume contrast-enhanced T1 MRIs suitable for volume measurements were available for 24 of 30 patients. Amongst the 6 patients lacking a scan of sufficient quality, there were 3 PN IDH1
mut
, 2 CLAS-like and 1 MES-like tumor. The scans were analyzed three-dimensionally in Brainlab iPlan® Cranial planning software version 3.0 (Brainlab AG, Feldkirchen, Germany), on which the enhanced area was interpreted as vital tumor, and the non-enhanced central area of the tumor as necrosis. As a measurement for total tumor area the enhanced and the non-enhanced central area were added up.
IHC procedure
Sections were cut from FFPE tissue in series and IHC staining was performed as described previously [
18]. Sections were stained for Carbonic anhydrase IX (CAIX), CD34, Endoglin (ENG), collagen type IV alpha 1 (ColIV) and α-smooth muscle actin (α-SMA).
Morphometrical analyses
The microvascular density (MVD) was microscopically assessed using the Chalkley grid. The average number of vessels per mm2, average vessel area and vessel perimeter were determined using computer-assisted morphometry.
Histological evaluation
The expression levels of CAIX, CD34, ENG, ColIV and α-SMA were quantified using Aperio ImageScope software. Vital tumor tissue was delineated on every individual section and staining, and hypoxic tissue was delineated using the CAIX staining pattern as guidance. The positive pixel percentage was calculated by division of the surface area found positive (in pixels) by the total area of the vital, normoxic or hypoxic field (in pixels), a method previously applied by others [
19,
20]. Since staining for α-SMA and ColIV was specifically identified around vessels, an indication of the thickness of these layers was obtained by dividing the number of positive pixels by the total vessel perimeter of that sample.
Microfluidic cards
RNA was purified from 30 snap-frozen GBM biopsy samples and reverse transcribed to cDNA. Custom-designed Taqman array Micro Fluidic Cards (low-density array, Applied Biosystems, Foster City, CA) were used to obtain gene expression analyses of these samples by analysis on a ViiA™ 7 real-time sequence detection system (Applied Biosystems). Assays were included for 31 genes of interest and one endogenous control.
Statistical analysis
Statistical analyses were performed using SPSS software version 22.0 (SPSS, Chicago, IL) and visualized using Graphpad Prism version 5 (Graphpad Software Inc, San Diego, CA). Correlations were calculated by Spearman’s rho (SR). Normal distributions were tested by one-way ANOVA or t test and data not normally distributed was tested by a Kruskal–Wallis test or Mann–Whitney U test. The multiple group comparisons were followed up by either Tukey’s or Dunn’s post-hoc tests. P values < 0.05 were considered significant and in all cases exact two-sided P values were reported.
Supplementary Methods can be found online (Online Resource 1).
Discussion
Because of the highly vascularized nature of GBMs we hypothesized that anti-angiogenic therapy could have an effect in these tumors. The novel therapeutics that seemed promising in pre-clinical studies, unfortunately, only rarely produced similar results in clinical trials. For example Bevacizumab, a promising anti-angiogenic drug in the pre-clinical setting, only resulted in an improvement of progression-free survival and left the overall survival of patients unaltered in two large Phase 3 randomized clinical trials when it was added to standard chemotherapy for newly diagnosed GBM patients [
25,
26]. A slight improvement in overall survival was observed for Bevacizumab addition to chemotherapy (Lomustine) for recurrent GBM patients [
27], but it is evident that more extensive research is needed to identify patients that could potentially benefit from these types of treatments. Therefore, in this study we aimed to identify vascular abnormalities specific of the PN
IDH1
mut
, CLAS-like and MES-like subclasses of GBM.
In agreement with previous reports, we found that MES-like tumors are enriched for necrosis [
21]. As it is known that necrotic tissue is often surrounded by hypoxic areas [
22], and hypoxia is a potent inducer of angiogenesis, high angiogenic potential was expected from MES tumors. By performing extensive characterization of the vasculature we were able to establish that MES-like tumors have larger, but not more vessels than other subclasses. This contradicted our hypotheses, since active angiogenesis would likely have resulted in predominantly small, newly formed vessels. Previous studies have however shown that larger vessels do not necessarily indicate better oxygenation, as the oxygenation status is more dependent on blood flow rate and diffusion distance from the vessel to the cells [
28]. Therefore, an increased flow rate in these tumors could still have resulted in a lower oxygenation level in MES-like tumors. Reduced perfusion in the MES-like tumors thus provides a potential connection between the larger vessel size in this subclass and the increased rate of necrosis and hypoxia. In this regard it would be interesting to analyze vaso-occlusive events in GBM, which itself has a profound effect on blood flow and oxygenation of target tissues [
29].
Furthermore, neo-angiogenesis could also be reflected by a more abundant expression of endothelial markers and a lower maturation rate of vessels. These hypotheses were partly confirmed by the data from the TCGA data repository, as those indicated elevated expression levels of endothelial markers CD34 and ENG in MES tumors at the mRNA level. At the protein level we were unable to identify such differences. Against our expectation, we found that the vessel maturity markers α-SMA and ColIV were elevated at mRNA level in the MES-like subclass compared to the PN IDH1
mut
subclass, and a similar trend was also observed at protein level. Because we established through morphometry that MES-like tumors have larger vessels we then applied a correction for vessel size. This correction led to fading of the differences observed at protein level, implying that the vessel size explained (part of) the differences in expression and there is no true difference in basement membrane deposition or pericyte coverage around the vessels of GBM subclasses. We believe that although the corrections for the number of vessels and vessel perimeter introduced differences between the mRNA and protein analyses, they can provide very useful information. It was due to these corrections that we were able to elucidate that the increased expression level of basement membrane and pericyte markers could actually be explained by an increase of vascular size.
The current study did not reveal an association between the angiogenic signaling network and GBM subclasses. As outlined earlier, genetic aberrations linked to particular molecular subclasses were presumed to affect different angiogenic downstream targets, however the expression of these targets appeared not to be different between the subclasses. Surprisingly, the few upregulated angiogenic targets were actually highest in CLAS-like and not MES-like tumors.
Given the scarceness of pre-clinical data it is interesting that a clinical trial with anti-angiogenic therapy (Bevacizumab) has included an analysis for the molecular subtypes. In this trial standard of care was supplemented with Bevacizumab (anti-VEGFA treatment) and especially the
IDH1 wild-type PN tumors benefited from the addition of Bevacizumab [
10]. Since our selection of PN tumors was based on an
IDH1 mutation, we were unable to address this subset of PN tumors.
In conclusion, we report that although MES-like GBMs have larger vessels and PN IDH1
mut
tumors tend to possess less necrosis and hypoxia, no major differences in angiogenic signaling patterns were observed between GBM subclasses at the time of diagnosis. Although our study did not
functionally address neo-angiogenesis, by the use of vascularization patterns and signaling signatures as a measure for the outcome of neo-angiogenic processes, our findings challenge the concept of MES GBMs being more angiogenic. It can be presumed that the lack of correction in prior studies possibly has led to the erroneous expectation of increased angiogenic activity in the MES subclass. As such, our results do not support attempts to differentiate anti-angiogenic treatment according to the GBM molecular subtypes assessed in this study.
Acknowledgments
The work in this article was funded by (1) The Graduate School of Medical Sciences, BCN-BRAIN, UMCG to Ms. S. Conroy, and The Dutch Cancer Society (KWF, RUG-2014-7471) to Dr. W.F.A. den Dunnen.