Angiogenesis and radiological tumor growth in patients with glioblastoma

As previously described, patients were retrospectively selected from all newly diagnosed GBM patients ≥18 years of age operated at St Olavs Hospital – Trondheim University Hospital, Norway between January 2004 and May 2014 (262 patients) [10]. Selection criteria were ≥ 2 pretreatment T₁-weighted postcontrast magnetic resonance imaging (MRI) scans separated by ≥14 days and histopathologically verified GBMs according to the 2007 World Health Organization (WHO) Classification [25]. Exclusion criteria were non-contrast-enhancing tumors and gliomatosis cerebri (defined by radiological criteria [26]). In addition, four cases were excluded because of insufficient tissue amount or morphology for the MVD assessments, which left 102 patients eligible for further analyses.

The segmentation of tumor volumes and establishment of growth rates have previously been described in detail [10]. The volume segmentation was performed by ALS and controlled by EMB (a neuroradiologist) using the software BrainVoyager QX (Brain Innovation, Maastricht, The Netherlands). Both preoperative MRI scans were segmented for total tumor volumes, defined as the combined volume of the non-contrast-enhancing central part (i. e. necrosis) and the contrast-enhancing rim. In addition, the reproducibility of the tumor volume assessments has been assessed and concluded as satisfactory [10].

The fitness of different growth patterns based on the segmented tumor volumes and the time intervals between the scans, have previously been assessed [10]. The Gompertzian growth pattern (Fig. 1) was concluded as the most biologically reasonable growth pattern [10], and all tumors were assumed to follow this growth pattern. Since growth rates were highly dependent upon tumor volume [10], a point estimate (such as doubling time) would be a wrong representation of tumor growth. To account for this issue, we calculated an expected Gompertzian growth curve based on the volume data from 106 patients [10]. The curve dichotomized the patients into having tumors with a larger or smaller volume increase than expected from the curve (i.e. fast-growing or slow-growing tumors) (Fig. 1) [10]. These two groups have previously been shown to associate with survival of GBM patients [12]. In the current study, these groups were assessed for associations with the MVDs.

All routine hematoxylin-eosin (HE) sections for each case have previously been microscopically assessed by VEM (a medical research student) and controlled by SHT (an experienced neuropathologist) for the presence of 30 different histopathological features, which have previously been described in detail [13]. These features were assessed for associations with the same groups of growth, and thromboses, high cellular density, vascular density, and mitotic count were of interest to this study. Thromboses were defined as vascular structures partly or completely occluded with fibrin. The general cellular and vascular densities of viable tumor areas were subjectively graded as low, moderate, or high. For statistical analyses, the 2 lowest categories were merged, because very few cases were graded as low in both variables. Mitoses were counted in 10 high-power fields (HPFs) in areas of highest mitotic counts (hotspots).

From each patient, one representative tumor sample from formalin-fixed paraffin-embedded (FFPE) tissue were cut at 4 μm, dried, deparaffinized, and rehydrated. Of the 102 cases, 11 cases had sections from FFPE frozen tissue. We applied vWF (vWF, 1:2000, polyclonal rabbit, EnV+/HRP, Dako Denmark AS, Glostrup, Denmark) and CD105 (CD105, SN6h, 1:50, monoclonal mouse, LSAB/HRP, Dako Denmark AS, Glostrup, Denmark). Optimum antibody concentrations were determined by titrations. For CD105, antigen retrieval was achieved with proteolytic enzymes, endogenous peroxidase activity was quenched with peroxidase block, and sections were incubated with the primary antibody for 24 h at 4 °C with cover glass after pretreatment with serumfree proteinblock. Both endothelial markers had negative controls and positive internal controls.

MVDs of vWF and CD105 were assessed by VEM, who was blinded to the growth data. A Nikon Eclipse 80i microscope and a Nikon CFI 10×/22 grid at × 400 magnification (area within the grid equal to 0.059 mm²) were used for the MVD assessments. MVDs were assessed after the methods of Weider et al. [27] with some modifications. MVDs were computed as the mean count of vessels within the grid for three HPFs of highest vascular densities (i. e. hotspots). Hotspots were identified by scanning using × 4 and × 10 objectives on vWF sections; corresponding hotspots were identified on CD105 sections. Only areas with ≥50% of viable central tumor tissue were counted. Tissue edges and areas with excessive hemorrhage and/or desmoplasia were avoided. Any individually stained endothelial cell or vessel within or in contact with the grid were counted. Moreover, because of the heterogeneous morphology of GBM vessels [28], each lumen was counted for long branched vessels and glomeruloid tufts as described by Kraby et al. [29]. In addition, separate units of ≥2 staining endothelial cells within the same vascular structure were counted as one vascular unit. Altogether, one case was not assessed for vWF-MVD due to high background staining, and another case was not evaluated for CD105-MVD due to non-existent antigenicity.

The estimation of growth rates and curves were computed using R version 2.13.1 [10] and analyses involving histopathology and MVDs were performed using IBM SPSS Statistics 24. Statistical significance was defined as p < 0.05 without corrections for multiple testing. The correlation between vWF-MVD and CD105-MVD was assessed using the Spearman rank correlation test. Associations between both MVDs and histopathological features were assessed using Mann-Whitney U tests (categorical vs continuous) and Spearman rank correlation tests (continuous vs continuous). vWF-MVD and CD105-MVD were further investigated for associations with fast-growing tumors using Mann-Whitney U tests. Finally, CD105-MVD was included in a multivariable binary logistic regression model with the histopathological features previously found to be significantly associated with faster growth in the same patients (thromboses, high cellular density, and mitotic count) [13]. Mitotic count was excluded from the model due to significant associations with all other features included [13].

Of the 102 included patients, 69% (70 patients) were male. The mean age was 63 years, range 26–83 years old. All cases were immunopositive for glial fibrillary acidic protein (GFAP) and two for isocitrate dehydrogenase mutation (IDH1-R132H) [12]. Eighty-two patients (80%) were preoperatively treated with corticosteroids. The median tumor volume was 17.7 mL (range 0.05–146.45 mL) from the first MRI scans and 27.5 mL (range 0.98–243.52 mL) from the second MRI scans. The median interval between the scans was 22.5 days, range 14–98 days. Overall, 52 patients (51%) had fast-growing tumors, and patient characteristics within the growth groups have previously been reported [12].

We observed that both markers stained endothelial cells quite specifically. vWF had a strong and granular cytoplasmic staining pattern, whereas CD105 had a more even and occasionally weaker cytoplasmic stain (Fig. 2). Generally, little background staining was observed, except in a few vWF sections. The distributions of both vWF-MVD and CD105-MVD were skewed, and the median vWF-MVD was 15.5 per field (range 0.7–62.0) and the median CD105-MVD was 12.7 (range 0.7–50.0).

vWF-MVDs and CD105-MVDs were significantly correlated (p < 0.001, ρ = 0.92). A scatterplot showing the relationship is found in Fig. 3. The median CD105-MVD/vWF-MVD ratio was 0.91 with 95% confidence interval (0.87, 0.95).

Of the features previously found to significantly associate with faster growth in univariable analyses (thromboses, high cellular density, and mitotic count) [13], only mitotic count was significantly associated with any MVD in this study (Table 1). However, CD105-MVD tended to be higher in cases with present thromboses or high cellular density despite the non-significant associations. The same tendency was observed for vWF-MVD and thromboses (Table 1). Interestingly, both MVDs were significantly associated with subjectively evaluated high vascular densities on HE sections.

In this study, only CD105-MVD was significantly associated with faster tumor growth in the univariable analyses (Table 2). However, the ranges of both MVDs were quite wide within both fast and slow-growing tumors (Table 2, Fig. 4). Nevertheless, CD105-MVD was no longer significant in a multivariable model including thromboses and high cellular density, where the two latter features were significant independent predictors of faster growth (Table 3).

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.

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).

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].

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.