Aberrant glioblastoma neovascularization patterns and their correlation with DCE-MRI-derived parameters following temozolomide and bevacizumab treatment

Glioblastoma (GBM) is a highly angiogenic malignancy, and its abundant, aberrant neovascularization is closely related to the proliferation and invasion of tumor cells. However, anti-angiogenesis combined with standard radio-/chemo-therapy produces little improvement in treatment outcomes. Determining the reason for treatment failure is pivotal for GBM treatment. Here, histopathological analysis and dynamic contrast-enhanced MRI (DCE-MRI) were used to explore the effects of temozolomide (TMZ) and bevacizumab (BEV) on GBM neovascularization patterns in an orthotopic U87MG mouse model at 1, 3 and 6 days after treatment. We found that the amount of vascular mimicry (VM) significantly increased 6 days after BEV treatment. TMZ inhibited neovascularization at an early stage, but the microvessel density (MVD) and transfer coefficient (Ktrans) derived from DCE-MRI increased 6 days after treatment. TMZ and BEV combination therapy slightly prolonged the inhibitory effect on tumor microvessels. Sprouting angiogenesis was positively correlated with Ktrans in all treatment groups. The increase in VM after BEV administration and the increase in MVD and Ktrans after TMZ administration may be responsible for treatment resistance. Ktrans holds great potential as an imaging biomarker for indicating the variation in sprouting angiogenesis during drug treatment for GBM.

impairs its therapeutic effect, and the mechanisms of TMZ resistance are still not clearly understood. Meanwhile, TMZ has been reported to inhibit tumor angiogenesis 15 . However, it is not clear which types of neovascularization patterns are sensitive to TMZ and BEV treatment.
Pre-clinical studies are an important approach to explore tumor neovascularization patterns. For example, vascular co-option has been demonstrated in a rat C6 glioma model 16 . Vasculogenesis has been further demonstrated in the angiogenic, defective tumor resistant Id-mutant mouse model 17 . Glioblastoma-endothelial cell transdifferentiation has been shown in orthotopic GBM mice models 18 . Using orthotopic rat C6 glioma models, our group discovered a new mosaic pattern of glioma vascularization 5 and confirmed vascular co-option, sprouting angiogenesis, intussusceptive microvascular growth (IMG) and vascular mimicry in the tumor region. We also demonstrated that neovascularization patterns varied along with tumor development and that dynamic contrast-enhanced MRI (DCE-MRI) can be used to evaluate neovascularization patterns in a glioma model 19 .
Vascular parameters can be dynamically evaluated by MRI in vivo. Dynamic susceptibility contrast MRI (DSC-MRI) can assess tumor vessel perfusion, and DCE-MRI can be used to evaluate vascular permeability 20 . DCE-MRI measures the MR signal to determine the change in concentration of the contrast agent over time within a field of view (FOV), and its quantitative parameters can be obtained through different pharmacokinetic models. For example, K trans reflects vessel permeability, V p reflects plasma volume, and K ep reflects regurgitation of contrast media 21 . K trans has been reported to be strongly correlated with glioma vascular permeability and microvessel density (MVD) after anti-angiogenesis therapy 22,23 . Therefore, we used DCE-MRI to monitor the effect of anti-angiogenesis therapy in this study.
Recently, many studies have demonstrated that different neovascularization patterns are mediated by corresponding signaling pathways 24,25 with varying sensitivities to anti-angiogenic drugs 26,27 . However, few studies have examined the change in neovascularization patterns after treatment for GBM. In our study, orthotopic U87MG glioblastoma mouse models were administered TMZ, BEV or a combination of BEV and TMZ, and then DCE-MRI scanning and histopathological analysis were performed to investigate the changes in neovascularization patterns and MVD in the tumor region. We aimed to determine the relationship between neovascularization patterns and resistance to drug treatment and investigated MRI biomarkers that indicate changes in neovascularization patterns during treatment.

Results
MRI features and neovascularization patterns of a U87 GBM model. GBM presented as a quasi-circular high signal in the right cerebral hemisphere in T2-weighted images. The signal was homogeneous, and mild edema was observed around the tumor mass. The K trans map derived from DCE-MRI showed high permeability in tumor tissues, especially at the tumor margin, indicated by bright points (Fig. 1). Four types of neovascularization patterns were identified by pathological staining of the tumor region, including vascular co-option, sprouting angiogenesis, IMG and vascular mimicry.
Histopathological results. The reproducibility of the histopathological results were assessed using the intraclass correlation coefficient (ICC = 0.95). Different treatment protocols had varying effects on tumor neovascularization patterns. In the BEV group, BEV had no effect on the MVD or neovascularization patterns 1 day after treatment. The amount of sprouting angiogenesis and IMG began to decrease 3 days after treatment compared with control groups (P < 0.001, P = 0.002, respectively). The MVD and the amount of IMG were both decreased compared with control group (P < 0.001, P = 0.001, respectively) and an actual decrease was observed in the number of co-option and sprouting angiogenesis 6 days after treatment (P < 0.001, P < 0.001, respectively) ( Fig. 2), whereas the amount of vascular mimicry was significantly increased (P < 0.001) (Fig. 3). In the TMZ group, TMZ inhibited tumor neovascularization 3 days after treatment. The amounts of sprouting angiogenesis, IMG and MVD were all significantly decreased compared with control groups (P = 0.001, P < 0.001, P < 0.001, respectively) and an actual decrease was observed in the number of vascular co-option (P = 0.001) 3 days after treatment. IMG was the most sensitive neovascularization pattern to TMZ and decreased 1 day after treatment compared with control groups (P = 0.001). However, 6 days after treatment, the MVD was significantly increased compared to the control group (P = 0.003), and no significant differences were found in the amounts of vascular co-option, sprouting angiogenesis, IMG and vascular mimicry between the experimental group and the control group (Fig. 4). In the BEV and TMZ combined group, the combination of BEV and TMZ significantly reduced the MVD and the amount of vascular co-option, sprouting angiogenesis and IMG 1 day (P < 0.001, P < 0.001, P < 0.001, P = 0.010, respectively) and 3 days (P = 0.003, P < 0.001, P < 0.001, P = 0.025, respectively) after treatment compared with control groups, but the amount of vascular mimicry (VM) was not different between the experimental and control groups. However, 6 days after treatment, the only decrease was observed in sprouting angiogenesis (P < 0.001) (Fig. 5).

DCE-MRI and correlation analysis.
The reproducibility of the DCE-MRI results were assessed using the intraclass correlation coefficient (ICC = 0.94). In the BEV treatment group, the K trans significantly decreased from day 1 through day 6 after treatment compared with control groups (P < 0.001 at 1, 3 and 6 days, respectively). In the TMZ treatment group, K trans significantly decreased 3 days after treatment but increased on the 6th day after treatment compared with control groups (P < 0.001, P = 0.007, respectively). In the BEV and TMZ combination treatment group, the variation in K trans was similar to that in the BEV treatment group. K trans was significantly reduced from day 1 to day 6 after treatment compared to the control groups (P = 0.022, P < 0.001, P < 0.001 at 1, 3 and 6 days, respectively) ( Fig. 6). Spearman correlation analysis showed that K trans was positively correlated with sprouting angiogenesis in all experimental groups after drugs administration and had no correlation with MVD, co-option, IMG, or VM (All the P values < 0.001) (Fig. 7).

Discussion
GBM is a highly angiogenic malignancy, and one of the characteristics of GBM is abundant and aberrant neovascularization. These new blood vessels provide nutrition and oxygen for tumor growth and are also closely related to the proliferation and invasion of tumor cells 7 . However, the failure of anti-angiogenesis therapy and the resistance to TMZ treatment in patients with GBM highlight clinical dilemma that motivated us to explore a novel perspective for the treatment of GBM. In this study, we found that TMZ could inhibit neovascularization at an early stage but subsequently led to an increase in MVD and K trans . BEV could increase the amount of VM 6 days after treatment, which could be responsible for the failure of anti-angiogenesis and standard TMZ treatment in patients with GBM. In addition, we found that sprouting angiogenesis was positively correlated with K trans in all treatment groups, indicating that K trans holds great potential as an imaging biomarker to indicate the variation in sprouting angiogenesis during drug treatment for GBM.
We found that TMZ suppressed the neovascularization in orthotopic U87MG GBM mouse models, which was consistent with the research reported by Hjalmar Kurzen 28 . The MVD and the amount of vascular co-option, sprouting angiogenesis and IMG were obviously decreased on the 3rd day after treatment in the tumor region. However, on the 6th day after TMZ administration, the amounts of vascular co-option, sprouting angiogenesis, IMG and vascular mimicry were not different between the experimental and control groups. The MVD and K trans even increased, suggesting that exuberant angiogenesis in GBM might be a crucial cause of chemo-therapy resistance and also demonstrating the necessity of anti-angiogenesis in the treatment of GBM.
BEV can inhibit neovascularization by competitively binding to VEGF and blocking its biological activity. Previous studies have shown that anti-angiogenesis treatment increases tumor cell invasion and vascular co-option [29][30][31] . In the present study, BEV began to reduce the amount of sprouting angiogenesis and IMG 3 days after treatment. The MVD and the amount of co-option, sprouting angiogenesis and IMG were all decreased on the 6th day after treatment, whereas the amount of vascular mimicry was significantly increased, indicating that vascular mimicry may precede tumor cell invasion and vascular co-option in response to anti-VEGF therapy. Vascular mimicry is associated with the prognosis of laryngeal cancer 32 , breast cancer 33 and glioblastoma 34 , suggesting that the increase in vascular mimicry is an important reason for the failure of BEV treatment at an early stage.
By dynamically monitoring the distribution of contrast media in the region of the tumor blood vessels and the tissue space, DCE-MRI can directly or indirectly reflect the structure and function of tumor blood vessels 21 . DCE-MRI has been widely used for early diagnosis 35 and evaluating therapeutic efficacy 36 . K trans , a major parameter of DCE-MRI, reflects vascular permeability and has been used as one of the main indicators for the clinical evaluation of anti-tumor drugs 37 . In our study, we utilized DCE-MRI to evaluate the effect of TMZ and BEV on GBM neovascularization patterns. Numerous studies have demonstrated that the breakdown of normal vascular structures by tumor cells is not only an important process to increase blood supply but also an important indicator of tumor invasion and metastasis 7 . Sprouting angiogenesis is a major modality for tumors to generate new vessels from preexisting ones. We found that sprouting angiogenesis was positively correlated with K trans in all treatment groups, suggesting that K trans is a promising imaging biomarker for indicating changes in sprouting angiogenesis after treatments.   In this study, we found that K trans significantly decreased from day 1 to day 6 after BEV treatment compared with control groups. According to previous reports, normalization of vascular abnormalities emerged after BEV administration and could improve the effect of radio-/chemo-therapy 38 . Our results are consistent with the hypothesis that K trans is an imaging biomarker of vascular normalization.
The addition of BEV to standard radio-/chemo-therapy after surgery has been reported to relieve clinical symptoms and prolong progression-free survival but not overall survival in patients with GBM 39 . In our observations, the greatest anti-angiogenesis effect in orthotopic U87MG GBM mouse models was achieved 3 days after combined TMZ and BEV administration (TMZ treatment was administered 24 hours after the administration of BEV), suggesting that an optimization treatment cycle could be determined for clinical management to achieve the greatest inhibitory effect on tumor neovascularization and growth.
Nevertheless, our study has some limitations. Patient-specific orthotopic GBM xenograft models should be used to accurately reflect the histological and genetic characteristics of the primary tumor and mimic the treatment response of parental GBM 40 .
In conclusion, we found that TMZ could stimulate neovascularization 6 days after treatment, and the amount of VM was increased 6 days after BEV treatment, which may be responsible for treatment failure. K trans can non-invasively indicate the variation in sprouting angiogenesis and reflect the function of tumor microvessels during anti-angiogenic treatment.  Anesthetized mice were mounted dorsally with their heads fixed in a stereotaxic apparatus (RWD Life Science, Shenzhen, China). A 5.0 µL cell suspension containing 5 × 10 5 U87MG cells was injected into the brain using a 20.0 µL microsyringe. The microsyringe was fixed stereotactically to vertically penetrate 4.0 mm into the brain mass, 2.0 mm to the right of the midline suture and 2.0 mm posterior to the outer canthus and was then retracted 0.5 mm. The cell suspension was then injected slowly over a period of 5 min. After injection, the microsyringe was held in place for 8 min and then retracted.
BEV and TMZ administration. Tumor-bearing mice were randomly divided into three groups. Drug administration was performed on day 21 after cell inoculation. MRI scanning were performed prior to treatment (Supplementary Dataset 1). The experimental groups were orally treated with 50 mg/kg TMZ (Merck, NJ, USA) for 5 consecutive days, intravenously administered 15 mg/kg BEV (Avastin, Roche Pharma Ltd, Mannheim, Germany), or administered a combination of TMZ and BEV (TMZ treatment occurred 24 hours after the administration of BEV). The control groups were administered saline in the same way with corresponding experimental groups. The experimental groups (5 mice) and corresponding control groups (6 mice) were examined at 1, 3 and 6 days after treatment. The process is shown in Fig. 8. MRI data processing. DCE-MRI data were processed after being imported into Omni Kinetic (GE Healthcare, Shanghai, China). K trans pseudo-color images were calculated following a "reference region" model proposed by Cárdenas-Rodríguez 41 . Then, the K trans of the tumor was calculated using the hot-spot method 42 : Five regions of interest (ROIs) with higher K trans values were chosen in the tumor area, and the K trans value of the tumor was represented by the average K trans value of the five ROIs. K trans values were measured by two experienced colleagues with MRI data analysis experience who were blinded to the group information and pathological results.
Histopathological analysis. After MRI scanning, tumor-bearing mice were immediately sacrificed using an overdose of anesthesia. Brains were harvested, and a fraction of tumor tissue was randomly selected and divided into pieces of approximately 1 mm 3 and fixed with 2.5% glutaraldehyde for transmission electron microscope analysis according to the standard procedures described previously 5 . The other tumor tissues were fixed with 4% paraformaldehyde and embedded with paraffin. Paraffin blocks were sectioned to perform H&E and immunohistochemical staining. The antibodies used were raised in rabbits against the following antigens in humans: CD34, glial fibrillary acidic protein (GFAP) and Tenascin-C (Abcam, Cambridge, UK). Two-micron-thick serial sections were used for immunohistochemical staining after dewaxing in xylene. Antigen retrieval was performed in a boiling citric acid solution (pH 6.0) for 2 min, and then slices were washed with PBS after natural cooling. H 2 O 2 (10%) and goat serum were used to block endogenous peroxidase activity and nonspecific antigens, respectively. Each slice was incubated overnight with solution containing the primary antibody at 4 °C. Specimens were then washed with PBS and incubated with HRP-conjugated goat anti-rabbit secondary antibodies at 37 °C for 30 min. DAB was used to visualize antigens. Dual CD34-periodic acid-Schiff (PAS) staining was performed as described previously 43 after CD34 immunohistochemical staining. Then, slides were counterstained with Mayer's hematoxylin and cover slipped with a permanent mounting medium.
Histopathological quantification. Histopathological quantification was performed by two colleagues who were experienced in histopathological analysis and blinded to the group information. Vascular co-option consisted of GFAP-positive glioma cells circumfusing CD34-positive tubes to form vascular niches 16 . Five fields with maximum co-option were selected under a 100× light microscope. The amount of vascular co-option in the tumor was represented by the average amount of vascular co-option in the five fields.
CD34-positive folds across the lumens of newly formed blood vessels were the key feature of intussusceptive microvascular growth 44 , which was quantified as the average number of folds in 10 lumens with more folds in the tumor area.
Specialized Tenascin-C-positive endothelial cells mediated sprouting angiogenesis 45,46 . Five fields with maximum sprouting angiogenesis were selected using a 200x light microscope, and the number of Tenascin-C-positive stalk cells was counted. Sprouting angiogenesis was quantified as the average number of Tenascin-C-positive stalk cells in the 5 fields.
Vascular mimicry was characterized by lumen-like structures with PAS-positive and CD34-negative basal membrane tissues 47 . We quantified vascular mimicry in the whole tumor region.
Five fields with maximum CD34-positive lumens were selected using a 200X light microscope. The MVD was quantified as the average number of CD34-positive lumens in the 5 fields.
Statistical analysis. SPSS 19.0 (IBM, Armonk, NY, USA) was used for the statistical analysis, and data are represented as the means ± SD. Image features and pathological results reproducibility were assessed using the intraclass correlation coefficient (ICC). Two sample t-tests were used to compare inter-group differences. Spearman correlation analysis was used to compare the correlation between K trans and neovascularization patterns. A value of P < 0.05 was considered significant.