A vascularized tumoroid model for human glioblastoma angiogenesis

Glioblastoma (GBM) angiogenesis is critical for tumor growth and recurrence, making it a compelling therapeutic target. Here, a disease-relevant, vascularized tumoroid in vitro model with stem-like features and stromal surrounds is reported. The model is used to recapitulate how individual components of the GBM’s complex brain microenvironment such as hypoxia, vasculature-related stromal cells and growth factors support GBM angiogenesis. It is scalable, tractable, cost-effective and can be used with biologically-derived or biomimetic matrices. Patient-derived primary GBM cells are found to closely participate in blood vessel formation in contrast to a GBM cell line containing differentiated cells. Exogenous growth factors amplify this effect under normoxia but not at hypoxia suggesting that a significant amount of growth factors is already being produced under hypoxic conditions. Under hypoxia, primary GBM cells strongly co-localize with umbilical vein endothelial cells to form sprouting vascular networks, which has been reported to occur in vivo. These findings demonstrate that our 3D tumoroid in vitro model exhibits biomimetic attributes that may permit its use as a preclinical model in studying microenvironment cues of tumor angiogenesis.


Results and discussion
The vascularized tumoroid model is based on spheroids, in which HUVEC and patient-derived primary GBM are in direct contact with each other, supported by HDF embedded in a fibrin gel (Fig. 1A). The NCH82 GBM cell line is also used. Variants of the model include GBM-only or HUVEC-only spheroids surrounded by HDF, GBM or no cells in the fibrin gel. It should be noted that direct comparison between the two cell lines is not possible because different types of culture medium were used. Graphical representations of (i) the total vessel area and vessel length for HUVEC spheroids surrounded by HDF, with or without additional growth factors and (ii) the total spheroid area for NCH82 or primary GBM spheroids. Data are presented as the mean ± standard deviation; n = 3; *p < 0.05, **p < 0.01, ***p < 0.001. Subscript S denotes cells seeded in the same spheroid. www.nature.com/scientificreports/ Sprouting of HUVEC spheroids. The first challenge towards the development of the vascularized tumoroid model is to provide a suitable environment for the HUVEC spheroids to form capillary sprouts. Angiogenesis is a complex process involving dynamic interactions between the endothelial and stromal cells, the production of angiogenic signalling factors, and is affected by changes in the surrounding ECM (e.g. hypoxic conditions). Fibrin was chosen as the supporting matrix for our experiments because it can be derived directly from the patients' blood 22 . Variation in the ECM composition is a hallmark of tumor stroma, with increased fibrin deposition associated with tumor angiogenesis 23 . Pro-angiogenic factors, such as VEGF and bFGF, have a high affinity for fibrin through the heparin binding domain 24 . Fibrin also triggers the production of other basement membrane proteins that are strongly linked to angiogenesis such as laminin and collagen IV 25 .
In the absence of a supporting cell type, endothelial cells cultured in spheroids within a fibrin gel undergo a migratory response instead of forming vasculature ( Figure S1). In angiogenesis models, a range of supporting cells can be used including human dermal fibroblasts, mesenchymal stromal cells and pericytes. In this study, we used the well-known method of culturing endothelial cells with human dermal fibroblasts, as they undertake significant collagen deposition and produce pro-angiogenic growth factors enabling capillary formation 26 . To this end, HUVEC spheroids are cultured with HDF single cells (at 10 6 cells/mL) within a 7.5 mg/mL fibrin gel to form angiogenic sprouts. Experiments are performed under normoxia (20% O 2 ) or hypoxia (1% O 2 ), with or without additional pro-angiogenic growth factors (200 ng/mL VEGF and bFGF). Figure 1B(i) shows a typical HUVEC spheroid at day 3 after seeding (additional timepoints are shown in Figure S2). In all cases the HUVEC display an angiogenic-like response, sprouting radially outwards forming capillaries. The addition of exogenous growth factors, as well as normoxia, induce a stronger angiogenic response compared to hypoxia. Figure 1C(i) shows the increase in vessel area and total vessel length of the produced sprouts at day 3.
Invasion of GBM spheroids. Before incorporating the tumor cells into the sprouting spheroids, the invasion capability of GBM-only spheroids embedded in a fibrin gel was examined for primary and NCH82 GBM cells under hypoxic and normoxic conditions. Figure 1B(ii) shows fluorescence images of the GBM spheroids at day 3. The corresponding total spheroid areas are shown in Fig. 1C(ii). It can be seen that hypoxia significantly increases GBM spheroid area in both primary and NCH82 cells which has been previously reported 27 . This effect appears to be stronger for primary cells compared to NCH82 cells.
Vascularized tumoroid model. The vascularized tumoroid model was created by applying vascularization techniques to tumor spheroids which were presented in the previous section. This model comprises of 1000-cell spheroids, consisting of HUVEC and GBM cells at a ratio of 3:1, in a 7.5 mg/mL fibrin gel containing 10 6 HDF cells/mL. Both primary and NCH82 GBM cells were used. The primary GBM cells were cultured in serum-free medium that preserves the subpopulation of glioblastoma stem-like cells as demonstrated in previous work 28 using the same cells, whereas the NCH82 GBM cell line was not, resulting in differentiated cells. Culturing GBM cells under serum-free conditions with hEGF and hbFGF has been reported 29,30 to tentatively preserve the characteristics of the primary brain tumors. Figure 1B(iii) shows fluorescence images of the GBM : HUVEC spheroids embedded in a 7.5 mg/mL fibrin gel containing HDF single cells, at day 3 under hypoxic and normoxic conditions. This tri-culture method was found to trigger robust HUVEC angiogenesis creating lumenised capillaries that sprout radially outwards into the surrounding fibrin gel. As in the case of HUVEC-only spheroids, angiogenic sprouting is higher at 20% O 2 compared to 1% O 2 . Moreover, the GBM cells behave differently depending on whether they are primary or not. NCH82 GBM cells invade into the gel without interacting strongly with the endothelial cells (this is more evident at longer timepoints as shown in Figure S3). On the other hand, primary GBM cells integrate more into the vasculature behaving similarly to the HUVEC. This phenomenon will be explored further in the following sections.
GBM promotes angiogenesis. In order to investigate the effect of glioblastoma cells on the sprouting of endothelial cells, the full tumoroid model (GBM : HUVEC spheroid surrounded by HDF single cells) or the exact model but without the GBM cells (HUVEC spheroid with HDF) were compared. Figure 2A depicts the angiogenic sprouting of HUVEC spheroids under various co-culture conditions at 1% and 20% O 2 . For each condition, the explant area, the average vessel length and the junction density are presented in Fig. 2B. The results show that in the presence of primary glioblastoma cells, endothelial cells produce vasculature of higher explant area but lower interconnectivity. In addition, an increase in the vessel length was observed under hypoxic conditions only (Fig. 2B). From the above, we hypothesize an invasive mode of angiogenesis for primary GBM cells. This suggests that the cells are producing tumor specific growth factors. The NCH82 GBM cell line interacts differently with the endothelial cells inducing sprouting within a smaller explant area while the average vessel length remains unaffected. As with the primary GBM, capillaries of lower interconnectivity are produced (Fig. 2B).
These results suggest a difference in angiogenic capabilities between primary and NCH82 GBM cells. Patientderived primary GBM cells have been cultured under conditions that retain their stem-like potential 28,31 whereas in the serum-grown NCH82 cell line the cells are differentiated. This is consistent with in vivo observations as tumors that have a high stem cell number are highly angiogenic 15,16 . Furthermore, the lower vascular interconnectivity observed in the presence of GBM cells is an important biomimetic feature as it is consistent with the high amount of dead ends present in the GBM vasculature compared to normal vasculature 32,33 .

GBM as supporting cells.
In a related experiment we tested the hypothesis that primary GBM cells can act as a supporting cell type for vessel formation, role taken up by the fibroblasts in the previous experiment. For this experiment, a 1000-cell HUVEC spheroid was cultured in a 7.5 mg/mL fibrin gel containing 10 6 cells/mL pri-  Fig. 3A for day 3. The number of HUVEC sprouts is shown in Fig. 3B. When primary GBM is used as a supporting cell, endothelial sprouting is more pronounced in hypoxia compared to normoxia (Fig. 3B). The opposite result is observed for HDF supporting cells ( Fig. 2A), demonstrating the supporting role primary, serum-free cultured GBM cells have especially under hypoxia even without direct intracellular contact. GBM cells are known to secrete pro-angiogenic factors under hypoxia, such as VEGF and bFGF 34,35 .
Furthermore, the addition of exogenous growth factors enhances this effect for cells cultured at normoxia but not at hypoxia suggesting that under hypoxia the GBM cells are already producing a significant amount of growth factors (Fig. 3B). In the absence of GBM cells and growth factors, HUVEC undergo a migratory response ( Figure S1) which suggests that GBM cells are indeed producing pro-angiogenic factors and these factors recruit cells that may contribute to vessel formation. Performing the same experiment using the NCH82 GBM cells,    www.nature.com/scientificreports/ endothelial cells are found to mostly invade the gel rather than sprout, particularly under normoxia, showing that these cells are not so capable of inducing HUVEC vascularization (Fig. 3A).

GBM angiogenic behavior.
Time-lapse images of the GBM vascularized tumoroid model depict the evolution of GBM : HUVEC spheroids with HDF single cells at 1% O 2 over 32 h starting at day 1 after seeding. In a feature unique to primary GBM vascularized tumoroids, tumor cells grow together with the endothelial cells forming sprouting networks. Moreover, as time-lapse imaging shows, some of the formed sprouts are composed of primary GBM cells (Fig. 4A). However, in NCH82 vascularised tumoroids, GBM cells break away from the primary mass and invade separately into the surrounding matrix (Fig. 4B). The co-localization between the tumor and the endothelial cells is quantified using the Pearson's correlation coefficient (Fig. 4C). The coefficient is steadily high ∼ 0.75 for primary GBM cells but it decreases from ∼ 0.6 to ∼ 0.2 for NCH82 GBM cells over 32 h. CD31 immunohistochemical staining of primary GBM : HUVEC vascularised tumoroids (Fig. 4D) showed angionenic sprouting, with cells other than HUVEC expressing CD31, suggesting that the primary GBM cells may have transdifferentiated into endothelial cells (Fig. 4D, arrow in the magnified views of the selected area (yellow box)). This effect was found to be more prominent under hypoxia which is consistent with what has been observed in vivo 36,37 .

Conclusion
A human, disease-relevant, in vitro vascularized tumoroid model has been developed to investigate microenvironment cues of glioblastoma (GBM) angiogenesis. The model comprises of GBM and endothelial cell (HUVEC) spheroids in a fibrin gel containing human dermal fibroblasts. It is easily scaled up, tractable, cost-effective, patient-specific, can employ naturally-derived and synthetic matrices, and is amenable to imaging and detailed analysis such as spheroid growth tracking. Although the model lacks perfusable vasculature and has no preexisting capillary bed or blood brain barrier, it is complex enough to recapitulate features of GBM angiogenesis, which have previously been observed in vivo. Patient-derived primary GBM cells were found to support angiogenic sprouting when cultured under conditions that preserve their stem-like potential, in contrast to a GBM cell line containing differentiated cells. Angiogenic sprouting was observed even in the absence of supportive fibroblasts. In the presence of exogenous growth factors, sprouting was enhanced under normoxia but not at hypoxia suggesting that hypoxic GBM cells are already producing a significant amount of growth factors. Primary GBM cells co-cultured with HUVEC in spheroids, produced vasculature of higher explant area and lower connectivity in comparison to HUVEC-only spheroids. Under hypoxia, primary GBM cells co-localize with HUVEC to form sprouting vascular networks. These findings demonstrate that our 3D tumoroid in vitro model exhibits biomimetic attributes. It therefore holds potential as a simple reductionist model for the development of anti-angiogenic treatments to GBM especially in view of the need to assess the preliminary effectiveness of new treatments before moving to more complex and lengthy efficacy in vivo models.
Multi-cellular spheroid formation. Co-culture spheroids were prepared using the hanging drop method.
A stock solution of methylcellulose was prepared by dissolving 1.2 g methycellulose powder (Sigma-Aldrich, UK) in 100 mL of DMEM. GBM and/or HUVEC cells were detached and suspended in 1:1 EGM to SFM medium containing 20% methycellulose (v/v). 20 µ L cell suspension drops each containing 1000 cells were pipetted onto a non-adherent petri dish (Greiner, UK). Co-culture spheroids were composed of HUVEC to GBM at a 3:1 ratio. HUVEC-only and GBM-only spheroids were also generated. The dish was turned upside down and incubated at 37 °C. The spheroids were harvested after 16 h.  Tumoroids were immunofluorescently stained with CD31 antibody. Gel constructs were fixed using 4% paraformaldehyde (PFA, Sigma-Aldrich, UK) for 1 hour followed by washes with phosphate-buffered saline (PBS, Sigma-Aldrich, UK). The constructs were subsequently permeabilised with 0.25% Triton X-100 (Sigma-Aldrich, UK) and blocked with 1% bovine serum albumin (BSA, Sigma-Aldrich, UK). Cells were incubated overnight with CD31 antibody (Abcam, UK) diluted at 1:100 followed by 2 h of incubation with 1:500 AlexaFluor 568 goat anti-mouse IgG (abcam, UK) diluted at 1:500. Washes with PBS followed both incubation periods.

Gel construct preparation.
Visualization and quantification of sprouting angiogenesis. Images of sprouting tumoroids were taken using epi-fluorescence microscopy (Zeiss Axio Observer Z1 inverted microscope with an ORCA-Flash4.0 camera). The microscope was equipped with an incubator regulating CO 2 (5%), O 2 (20% or 1%), temperature (37 °C) and humidity (95%) (Okolab, Italy). Time-lapse images were acquired every 20 min. The images were the result of the deconvolution of an image z-stack using ZEN 3.0 software 38 .
The area of the vessel networks that extended beyond the surface of the initial spheroid boundary was selected using ImageJ. This area was then analysed using AngioTool 39 . The software provided automatic measurements of: (i) the total vessel area, (ii) the total vessel length, (iii) the explant area (area of convex hull containing all the vessels), (iv) the average vessel length and (v) the junction density (total number of vessel junctions normalised by the area). A vessel was defined as the segment between two junction points or a junction point and an end point. The total spheroid area (area of spheroid including any sprouting) was quantified in ImageJ.
GBM co-localization along blood vessels was quantified by the Pearson's correlation coefficient in Coloc 2 40 (ImageJ plugin 41 ).
Statistical analysis. Data comparisons between two sets of data were performed in GraphPad Prism 8.4.3 software 42 . Data are displayed as mean ± standard deviation. Three spheroids were analysed out of each one of three independent experiments. The explant area of each spheroid ranged between 0.1 and 1.3 mm 2 . Two sets of data were compared using the Students t-test. The threshold for statistical significance was set at a value of *p < 0.05. **p < 0.01; ***p < 0.001. Accession codes. The datasets generated and analysed during the current study are available to download from Mendeley Data, V1, https:// doi. org/ 10. 17632/ kc64v 677tb.1.