In this study, CD105-MVD, and not vWF-MVD, was significantly associated with faster growth in the univariable analyses. However, the relation was lost when adjusted for the presence of thromboses and high cellular density in a multivariable model, where these two latter features were significant independent predictors of faster growth. Both MVDs associated significantly with mitotic count, but neither with the presence of thromboses nor high cellular density.
Biological reasons for why some GBMs grow faster than others are sparsely studied in human patients, which is mainly due to difficulties in acquiring in vivo pretreatment growth estimates [
10,
11]. In addition, most research on growth processes have been conducted on in vitro or animal models which fail to accurately imitate the unique micro-milieu of the human GBM [
30]. Moreover, by having preoperative growth as an outcome variable instead of overall survival, we avoid the effects of clinical factors found to be influential on survival, such as age at diagnosis, tumor size at diagnosis, Karnofsky performance status, comorbidity, and effects of treatment. Corticosteroid treatment was the only preoperative treatment received by our patients (82 patients); however, it was not significantly associated with tumor growth when corrected for tumor volume [
10], and associations between histopathology and growth were independent of such treatment [
13]. Altogether, these aspects of the study made it possible to study the biology of GBM growth as unaffected as clinically justifiable.
Endothelial markers and angiogenesis
The prognostic role of MVD measurements in glioblastomas is unclarified [
31‐
35]. However, a few studies have reported that vWF-MVD and CD105-MVD may predict the malignancy grade and prognosis of gliomas [
36,
37]. The finding that only CD105-MVD was significantly associated with growth in the univariable analyses, is in line with other univariable studies which have found more promising results for CD105-MVD than for CD31-MVD (a pan-endothelial marker) as prognostic markers of GBM [
33,
34]. These studies speculated the potential prognostic inferiority of pan-endothelial markers (i.e vWF, CD31, CD34) was caused by the additional staining of pre-existing angiogenically inactive vessels [
23]. In contrast, many studies have shown that CD105 predominantly stain proliferating endothelial cells [
22,
33,
38‐
46]. However, a few studies have observed CD105-positive vessels in normal [
47] and GBM-adjacent brain tissue [
48], and the marker needs further validation. Moreover, it has been shown that vWF sometimes fail to stain microvessels in both normal and neoplastic tissue [
46].
vWF-MVD and CD105-MVD were highly correlated (Fig.
3), with a higher median and upper range for vWF-MVD. In addition, the high CD105-MVD/vWF-MVD ratio suggests most vasculature of GBM are angiogenically active. Because the markers were counted in corresponding HPFs, a high correlation coefficient was expected. However, some of the differences in the MVDs could be caused by random variations in vascular densities of different sections and the granular staining of vWF (Fig.
2), which sometimes made it more difficult to distinguish separate vascular units than in CD105 sections.
Neovasculature and tumor growth
CD105-MVD was no longer significantly associated with preoperative tumor growth in a multivariable model with thromboses and high cellular density, where both latter features were significant independent predictors of faster growth (Table
3). However, reverse causation may also be possible: faster growth could lead to thromboses and high cellular density. Furthermore, we observed that fast-growing tumors could have quite low CD105-MVD scores and slow-growing quite high (Table
2, Fig.
4), which was in line with the finding that CD105-MVD explained very little of the variance in speed of growth (3%) in the univariable analysis (data not shown). Similar ranges of CD105-MVD were observed within the growth groups when patients with sparse tissue amounts (46 cases) were excluded (Additional file
1), and the weak association was thus unlikely a result of sampling errors. Altogether, our results suggest that vWF-MVD and CD105-MVD are not predictive of faster GBM growth.
There are several biological mechanisms which could potentially explain the inferiority of CD105-MVD as an independent predictor of tumor growth. One reason could be that tumors can create a surplus of or ineffective vasculature due to excessive angiogenic stimuli [
23], potentially leading to an overrepresentation of MVD counts. Excessive angiogenic stimuli may be caused by oncogenic mutations (known as hypoxia-independent angiogenesis [
49]). Other explanations could be that other mechanisms of glioma-associated neovascularization account for additionally needed vasculature in fast-growing tumors [
49], such as vascular co-option [
50], vasculogenesis [
51,
52], vascular mimicry (non-endothelial vasculature) [
53], and glioblastoma-endothelial cell transdifferentiation [
42]. Vascular mimicry is the process most likely to be overlooked by our methodology due to the lack of endothelial cells. In addition, the presence of vascular mimicry has been found to significantly predict higher glioma grades and worse prognosis [
53]. However, it is uncertain to which degree and how the different processes of neovascularization interact, and further studies are needed [
49].
In our previous study, we speculated that hypoxia initiated by thromboses facilitated growth through an induction of angiogenesis [
13]. However, the finding that the presence of thromboses was still a significant independent predictor of faster tumor growth when the degree of angiogenesis was not, suggests angiogenesis-independent mechanisms driven by hypoxia contribute to faster GBM growth. Such hypoxia induced mechanisms may act through other mechanisms of glioma-associated vascularization [
49], augmentation of proliferation [
54,
55], and initiation of invasiveness [
19,
55]. Increased invasiveness is one of the proposed mechanisms of resistance to anti-VEGF (bevacizumab) treatment [
55‐
59], and thus the lack of survival benefit in randomized trials [
20,
21]. Some studies even suggest GBM growth is possible without an induction of angiogenesis [
19,
60]. Such angiogenesis-independent growth may be described by the “go-or-grow” hypothesis, where tumor cells switch between two mutually exclusive phenotypes of either proliferative or invasive characteristics [
61,
62]. Hypoxia has been proposed to induce the switch to the invasive phenotype [
62]. In this way, hypoxic tumor cells migrate away from hypoxic areas and switch back to a proliferative phenotype when nutrients are adequate without an induction of angiogenesis [
60]. In addition, Sakariassen et al. [
19], discovered that xenotransplanted GBMs in nude rats could present as fatal diseases without signs of angiogenesis. Nevertheless, invading cells are unlikely to be captured as contrast enhancement without an induction of angiogenesis [
19,
59,
63], and are therefore unlikely to have been measured in our study. Additionally, the non-significant multivariable association for CD105-MVD could perhaps be caused by the nearly significant associations between CD105-MVD and thromboses and high cellularity (Table
1). Collectively, our findings support that angiogenesis-independent mechanisms driven by hypoxia contribute to faster GBM growth, which might explain the lack of survival benefit of anti-VEGF treatment.
As thromboses, high cellular density maintained its role as a significant independent predictor of faster growth in our study (Table
3). This finding substantiates our previous speculation that it is a better marker of high proliferation rates than high mitotic counts, because mitotic count has many potential sources of errors [
64], and higher counts were significantly associated with the presence of thromboses [
13] and increasing CD105-MVD counts (Table
1).
Microvessel methodology
So far, there is no standard method for quantification of MVD, however, initiatives on international standardizations have been made [
65]. Like in our study, most studies are based on the methods described by Weidner et al. [
27] with modifications: they count single positive cells and avoid areas of sclerosis, necrosis, and non-neoplastic tissue. However, few have specified their handling of glumeruloid tufts and longer vessels. We believe as Leon et al. [
36], that counting a glomeruloid tuft as one vascular unit might underestimate the angiogenic stimuli of the tumor. Furthermore, the subjective assessment of hotspots and interpretation of positive immunostaining give rise to problematic inter-observer variability, which has been reported as quite high for MVD assessments in GBMs [
66].
Even though we found significant associations with both MVDs and subjectively assessed high vascular densities on HE slides, the spreads of the MVDs were wide within and overlapping between the categories of vascular density (Table
1). These findings were in line with the fact that capillaries is known to be inconspicuous on HE slides [
28].
Strengths and limitations
Limitations of the assessments of tumor volumes on MRI scans, growth rates, and histopathological features have previously been described in detail [
10,
13]. The main strengths are the relatively large number of patients with a population based referral and the reproducibility assessment of tumor volumes [
10]. Potential biases are selection biases, preoperative steroid treatment, differences in diagnostic MRI scanners, different timing and administration of the contrast agent, tumor cells existing beyond the contrast enhancing rim [
67,
68], and sampling errors and inter-observer variability of the histopathological assessments. Additionally, our analyses were exploratory and should be validated in future studies.