GSCs are among the multiple malignant cellular subpopulations of GBM. GSCs share some characteristics with neural stem cells, like being able to self-renewal, express stemness markers (e.g., Nestin or Olig2) and undergo multilineage differentiation. They are known to be intrinsically resistant to radiotherapy and chemotherapy, contributing to intratumor heterogeneity, invasion and recurrence after treatment (Liu et al., 2006; Shi et al., 2018; Mazor et al., 2019; Bhaduri et al., 2020).

The intimate relationships between tumor vasculature and GSCs have been recognized and studied for more than a decade. In 2007, Calabrese et al. reported the presence of GSCs around tumor capillaries in patient samples, which focused the attention of researchers in the connection between GSCs and blood vessels (Calabrese et al., 2007). Since then, several studies found that the narrow interaction between vasculature and GSCs is essential to support GSCs’ tumor propagating ability. Moreover, GSCs themselves stimulate blood vessel development, which justify their residence mostly in perivascular niches (Brooks and Parrinello, 2017).

Tumor blood vessels and GSCs crosstalk can be established by direct cell-to-cell contact or the release of soluble factors. For example, tumor vasculature ECs assist GSCs self-renewal and maintenance through the expression of Notch ligands, specifically the jagged canonical Notch ligand 1 (JAG1) and delta like canonical Notch ligand 4 (DLL4), which interact with the GSCs Notch receptor (Zhu et al., 2011). In fact, the knockdown of tumor-associated JAG1 and DLL4 in ECs compromises GSCs-derived tumor development, which proves the relevance of Notch signaling in the tumorigenic capacity of GSCs (Zhu et al., 2011). Nitric oxide (NO) is another important regulator in this crosstalk (Charles et al., 2010). It is known that GBM-associated ECs express high levels of endothelial NO synthase (eNOS). NO released from the endothelial layer was shown to diffuse into adjacent GSCs, thus regulating the ability of these cells to form the tumor tissue in vitro and in vivo (Charles et al., 2010). Similarly, ECs-derived interleukin 8 (IL-8) has also been suggested as a mediator of GSCs proliferation and migration (Infanger et al., 2013). The Sonic hedgehog (Shh) protein is heavily produced and released from neovascular ECs and reactive astrocytes. Through binding to its receptor on GSCs surface, Shh activates the hedgehog pathway, known to maintain stem cells self-renewal (Becher et al., 2008; Yan et al., 2014). Over the years, evidence has also shown the involvement of fibroblast growth factor (FGF) in GSCs maintenance, supported by ECs. GBM cells in the presence of FGF can acquire a GSC phenotype, evidencing the contribution of tumor ECs in the maintenance of GSC pools (Fessler et al., 2015).

In turn, GSCs also contribute to the maintenance and expansion of the vascular compartment. GSCs can differentiate into tumor-derived ECs or mural-like ECs, increasing tumor vascularization (Soda et al., 2011; Scully et al., 2012). Moreover, GSCs sustain continuous angiogenesis by VEGF secretion (Bao et al., 2006). Therefore, it seems to be established a self-sustained positive feedback mechanism between GSCs and tumor-associated ECs - GSCs induce EC proliferation and survival, while the latter expands the perivascular niche, where GSCs reside, and ensure their maintenance (Figure 1).

GSCs are also present in hypoxic niches, far from perivascular niches. In the tumor bulk, hypoxia regulates stemness through hypoxia inducible factor 1-alpha (HIF-1α) and 2-alpha (HIF-2α) stabilization, an upstream transcription factor of self-renewal, dedifferentiation and pro-angiogenic genes (Li et al., 2009; Qiang et al., 2012). The knockdown of HIF-1α or HIF-2α in GSCs results in the loss of stem cell-like characteristics, affecting the expression and secretion of VEGF (Li et al., 2009; Méndez et al., 2010). Overall, this strongly supports the role of GSCs, both from perivascular and hypoxic niches, in tumor neovascularization.

As previously described, the GBM microenvironment harbors both malignant and non-malignant cells, including immune system cells. Various immune cell types can be found in different tumors, as regulatory T cells (Treg) or myeloid-derived suppressor cells (MDSCs) but, particularly in GBM case, microglia/macrophages are the most frequently present, representing about 30–40% of all tumor cell content (Vega et al., 2008; Shang et al., 2015; Zhang et al., 2016).

In the tumor site, macrophages can undergo different activation processes, acquiring a wide spectrum of phenotypes, depending on the surrounding environment (Fu et al., 2020). This activation is highly complex and dynamic, and evidence indicates that GBM-associated macrophages (GAMs) maintain a chronic inflammation in an initial phase, through the release of tumor necrosis factor alpha (TNFα), IL-6 and interferon gamma (INF-γ), creating a mutagenic environment favorable to tumor progression (Cassetta and Pollard, 2018). Then, GAMs preferentially differentiate into an anti-inflammatory phenotype (Mantovani et al., 2002). Anti-inflammatory or M2 macrophages also support tumor survival by suppressing anti-tumor immunity, namely cytotoxic T cells inflammatory responses (Hambardzumyan et al., 2016).

Tumor cells and tumor vasculature closely regulate GAMs, from their recruitment to their differentiation, maintenance and immunosuppressive role. Apparently, glioma cells initiate this process. They are major sources of SDF-1, glial cell derived neurotrophic factor (GDNF) and colony stimulating factor 1 (CSF-1), stimulating monocyte diapedesis and differentiation into M2 macrophages (Wang et al., 2012; Ku et al., 2013; Pyonteck et al., 2013; Sielska et al., 2013). In addition, endothelial cells release high levels of IL-6, which induces macrophage pro-tumoral phenotype, via HIF-2α activation (Wang et al., 2018). Therefore, GAMs concentrate mainly in tumors’ perivascular and perinecrotic niches (Zeiner et al., 2019).

Once in the TME, GAMs display a pro-angiogenic profile (Tamura et al., 2019). In fact, the most important angiogenesis inducer – VEGFA – is highly expressed by these cells. Also, GAMs’ enzymes, including matrix metalloproteinase 9 (MMP-9) and urokinase-type plasminogen activator (uTPA), drive ECM degradation, whereas other GAMs’ enzymes, such as thymidine phosphorylase (TP), drive pericyte and EC migration (Hotchkiss et al., 2003; Piao et al., 2005; Du et al., 2008; Riabov et al., 2014; Zhu et al., 2017). Therefore, macrophage recruitment and differentiation contribute to neovascularization and, consequently, for tumor growth and invasion. Overall, this suggests that tumor-associated ECs and GAMs support each other in the TME, and both support tumor growth (Figure 2).

As previously mentioned, GAMs impair cytotoxic T cell anti-tumor responses. GAMs present high levels of ectonucleotidase CD39, which cooperates with CD73 generating adenosine and, consequently, driving T cell immunity suppression (Takenaka et al., 2019). However, GBM is known as a “cold tumor,” meaning it presents low levels of T cells. This might happen because GAMs prevent T cell transmigration and tumor infiltration by interacting with ECs - pro-angiogenic factors derived from GAMs lower the expression of adhesion molecules on the surface of ECs. This mechanism, known as endothelial anergy, reduces the intercellular levels of adhesion molecule 1 (ICAM-1) and vascular adhesion molecule 1 (VCAM-1), essential for T cell transmigration and tumor accumulation, thus protecting the tumor from any potential immune attack (Wang et al., 2021) (Figure 2).

For a long time, ECM was seen as a purely structural tumor component. However, since the 1990s, evidence began to emerge that ECM is closely involved in cellular migration, differentiation, survival and invasion, contributing to tumor development and expansion (Bilozur and Hay, 1988; Koochekpour et al., 1995; Bouterfa et al., 1997). Healthy brain ECM has an absolutely unique composition, with relatively small amounts of proteins, as collagen or fibronectin, and greater content of glycosaminoglycans, such as hyaluronic acid, which may or may not be bound to proteins, forming proteoglycans (Novak and Kaye, 2000). This nature provides brain an amorphous soft structure, termed as perineural net (Mouw et al., 2014).

In contrast to what happens in healthy brain, brain tumors trigger profound alterations in the amount and deposition of ECM, its composition (particularly upregulation of collagen, fibrin and laminins), organization and post-translation modifications, which lead to the development of a less amorphous, denser and more rigid structure, necessary for disease establishment and progression (Belousov et al., 2019).

Both GBM malignant and stromal cells are important for the above-described ECM remodeling. Through an increased production of hyaluronic acid, laminin, collagen IV and VI or fibronectin, these cells create a more compact and denser ECM (Delpech et al., 1993; Bouterfa et al., 1999; Kawataki et al., 2007; Mammoto et al., 2013; Henke et al., 2019). In addition, these cells upregulate receptors and enzymes responsible for cell-matrix connections, which together with the tumor remodeled ECM, seems to facilitate tumor invasion, migration, angiogenesis and overall growth (Henke et al., 2019).

For instance, both tumor and tumor-related ECs are known to produce laminin α2 to a large extent (Kawataki et al., 2007). On its own, an increase in ECM laminin content, and consequently, in tumor matrix density, hampers drug diffusion and radiation penetration, protecting tumor against multiple treatments (Henke et al., 2019). In addition, direct contact between extracellular laminin and its integrin receptors, on GBM cells and GSC surface, is involved in tumor cell migration and radiotherapy resistance, respectively (Lathia et al., 2012; Ellert-Miklaszewska et al., 2020). Therefore, tumor-associated blood vessels enable tumor proliferation and resistance through their contribution to ECM abundance and stiffness.

According to previous studies, GBM core and margin stiffness is about ten times greater than normal brain tissue stiffness (1:0.1 kPa) (Miroshnikova et al., 2016). Besides creating a physical barrier, some studies reveal that ECM hardness dictates GBM cell behavior and angiogenesis. In response to stronger ECM forces, through mechanosensing and mechanotransduction processes, GBM cells change their size, shape, signaling pathways and density. For instance, in vitro studies showed that tumor cell compactness induces the production of collagen IV and VI, and the expression of lysyl oxidase, supporting an even greater ECM stiffness. This compactness, in turn, incites the expression of pro-angiogenic factors, leading to tumor vascularization (Mammoto et al., 2013).

As mentioned before, angiogenesis is a two-step process involving EC activation followed by endothelial cell sprouting, when ECs proliferate and migrate to assemble new blood vessels. During migration, ECs recognize and specifically bind to ECM ligands through integrin receptors, generating the necessary traction and cytoskeleton reorganization for this movement (Wang et al., 2005). Also, EC movement requires prior degradation of the vascular basal membrane and ECM remodeling through metallo, serine and cysteine proteases. uTPA, activator of proteolytic enzyme urokinase-type plasminogen (uTP), MMP9 and cathepsin B are highly expressed in GBM (Forsyth et al., 1999; Konduri et al., 2001; Rustamzadeh et al., 2003). ECM breakdown promoted by these proteases results in release of VEGF, FGF and epidermal growth factor (EGF), previously bound to ECM, and of matrix components that become free to interact with GBM cell integrins and activate MEK and PI3K signaling pathways, inducing the synthesis and/or activation of pro-angiogenic molecules (Lakka et al., 2005). Thus, tumor-induced ECM remodeling works as an angiogenesis activator signal, either by allowing EC sprouting, or by contributing to EC proliferation.

All these observations imply an obvious and close relationship between GBM blood vessels, ECM and tumor aggressiveness. Blood vessels contribute to the formation of regions of remodeled ECM, therefore promoting tumor proliferation, migration and invasion; on the other hand, the remodeled ECM supports continuous angiogenesis (Figure 3).

Spheroids are a group of cells that grow together in a spherical, non-adherent structure. Their multi-dimensional structure creates better conditions than classic 2D models to replicate physiologic-like cell-cell and cell-matrix interactions, pH gradients, oxygen availability and dynamic metabolic demands of tumors, representing the multiple in vivo tumor regions, from the senescent core to the proliferative periphery. Furthermore, several studies comparing tumor cell behavior in 2D and 3D systems revealed that spheroids better recapitulate the tumor genetic expression and chemotherapy resistance, often observed in vivo (Lin and Chang, 2008). From a technical point of view, they are inexpensive, easy to maintain and to manipulate (Lu and Stenzel, 2018). Thus, spheroid models have been extensively investigated and used to mimic human tumors, including GBM.

In the literature, the most frequently reported GBM spheroids comprise only one human or one mouse cell line and are used to test the tumor penetration ability and cytotoxic activity of novel therapies, including cutting-edge nanosystem-based therapies. For instance, in a recent work, Broekgaarden et al. used human derived U87MG spheroids and mouse derived F98 spheroids to test the tumor penetration and radiotherapy effectiveness of surface-functionalized gold nanoparticles (Broekgaarden et al., 2020). While these models are advantageous when compared with monolayered cell cultures, they still lack relevant TME elements, and do not mimic perivascular regions.

To overcome these weaknesses, other approaches are based on spheroids assembled with both tumor and endothelial cells, subsequently incorporated into hydrogels or biofilms made of Matrigel or human derived ECM components, as collagen, hyaluronic acid or fibrin. These systems allow better replication of the natural tumor tissue organization (Pan et al., 2019). Hydrogels generate a native tumor-like environment, allowing EC sprouting and tumor cell invasion, critical for brain tumor angiogenesis research (Rustad et al., 2012). In line with this, spheroids of GBM primary cells and EC embedded in ECM gels were recently reported (Table 1) (McCoy et al., 2019; Tatla et al., 2021b). In these models both EC and GBM cell migration was observed, which confirms models are appropriate for angiogenesis and GCS transdifferentiation study (McCoy et al., 2019; Tatla et al., 2021b). Through biochemical analysis of their spheroid model, McCoy et al. found that endothelial cells stimulate GBM cell migration, via IL-8 secretion. Furthermore, the model developed by McCoy et al. was useful to test potential GCS transdifferentiation inhibitors (McCoy et al., 2019).

Despite the benefits of these spherical models, they are limited by lack of the blood flow shear stress, which Plays an important role in in vivo neovascularization

Some researchers described the association of homogeneous tumor spheroids with BBB models using Transwell® systems, where both are separated by a permeable membrane. For drugs that need to cross the BBB before tumor internalization, these models allow BBB permeability and tumor penetration/toxicity testing in a more biologically relevant manner, but they do not represent the GBM vascularized tissue (Belhadj et al., 2017; Marino et al., 2019; Sherman and Rossi, 2019). Nevertheless, in vitro studies conducted with BBB models co-cultured with GBM cell monolayers in Transwell® systems showed that tumor cells drive EC uncontrolled proliferation and disorganization, like the native focal sites of BBB disruption in GBM scenarios (Mendes et al., 2015). Therefore, there is the possibility that these alterations also occur in BBB models co-cultured with GBM spheroids, mimicking tumor and tumor-associated BBB separately, but more accurately.

Cell cultures supported by biologically relevant matrices can create scaffold-based in vitro models, that attempt to mimic tumor biophysical environment. Given the previously described GBM ECM composition, Matrigel, collagen, fibrin and hyaluronic acid are the most frequently employed matrices to produce these models. In fact, studies reported that scaffold biophysical features, as stiffness, affects GBM stemness and, consequently, GBM chemotherapy resistance (Palamà et al., 2018; Xiao et al., 2020). So, these biomaterials mechanical properties have been manipulated to condition cellular behavior and create biomimetic models, including vascularized GBM models. Despite all the efforts made to capture tumors biophysical environment, this remains a great challenge. Multiple variables as porosity, polymer concentration, matrix elasticity must be simultaneously controlled to mimic natural tumor ECM (Vasudevan et al., 2020).

One of the first materials used as scaffold for in vitro models was Matrigel. In 1988, Kubota et al. found that human umbilical vein endothelial cells (HUVECs) on Matrigel attach and migrate towards each other forming long and complete capillary-like structures when supplemented with bovine calf serum and FGF, or only short and incomplete structures without any supplementation (Kubota et al., 1988). Because this capillary-like structures formation under certain conditions mimics the in vivo angiogenic process, Matrigel-based tubular networks assays emerged as useful tools to identify angiogenic signaling cascades and to assess new anti-angiogenic therapeutics (Benton et al., 2014).

Matrigel has also shown utility in the study of GSC transdifferentiation into ECs. A tubular network assay with patient-derived GBM cells cultured on Matrigel was used to characterize GSCs and verify their vascular development ability (Ke et al., 2020). In this case, patient cells were able to transdifferentiate into ECs and further form tubular structures. Using the same model, researchers also tested drugs capable of inhibiting this transdifferentiation process (Ke et al., 2020). Furthermore, patient-derived GBM cells cultured on Matrigel have also been used to study GBM recurrence and resistance mechanisms after radiotherapy. Apparently, radiotherapy enhances TDECs ability to form tubular structures in Matrigel, without increasing their proliferation rate (Deshors et al., 2019). Therefore, and as mentioned before, the increased vascularization capacity of TDECs may be one the mechanisms explaining GBM resistance and recurrence after radiotherapy. Considering these systems’ power to study angiogenesis, GSC transdifferentiation and to test novel neovascularization inhibitors, we decided to include them in this section even though they do not represent vascularized tumor tissue per se.

In the last decade, 3D bioprinting emerged has one of the strategies with greatest potential for in vitro models’ development. In general, this technique consists of bioink printing, layer-by-layer, allowing the establishment of complex 3D living systems (Han et al., 2020). There are two bioink subtypes: one includes living cells within an exogenous biomaterial matrix, both printed simultaneously, and the other includes only living cells, which are printed within a cast structure and then produce their own ECM (Dai et al., 2016). In 3D-bioprinted GBM models, only the first bioink type has been used.

Printed cells usually include cancer cell lines or patient-derived tumor cells, ECs, fibroblasts or immune cells, while alginate, gelatin, polyethyleneglycol (PEG) methacrylate or decellularized ECM comprise the models’ matrix. These models are usually very reproducible and adjustable since their production is entirely dependent on computer programs and no human manipulation is directly involved (Tang et al., 2021). In contrast with 2D cell cultures, 3D-bioprinted models maintain in vivo stemness properties, gene expression and vascularization tendency. Furthermore, they allow the manipulation of a specific factor to study its role in tumor biology (Ruiz-Garcia et al., 2020).

Recently, 3D bioprinting has revealed its applicability and usefulness in the neuro-oncology field, specifically in the assembly of GBM in vitro models to study neovascularization. Both GBM cells and ECs bioprinted within ECM-like materials sprout, forming tubular structures (Figure 4) (Table 1) (Han et al., 2020; Tang et al., 2021). Moreover, pro-angiogenic genes expression and cells’ sprouting is regulated by matrix biophysical properties, such as stiffness (Tang et al., 2021). Therefore, these models mimic in vivo angiogenesis and GSCs transdifferentiation.

In vivo behavior of tumor cells is also maintained in vascularized bioprinted models, in contrast to what is observed in other models. In vascularized bioprinted models, tumor cells are more invasive and resistant to therapy, which might be a result of vascular derived signaling cues (Figure 5) (Table 1) (Yi et al., 2019; Ozturk et al., 2020). Thereby, these models are useful in drug research and they may eventually help guiding clinical decisions.

Although none of the developed GBM vascularized models has been explored to study cell-cell interaction, we believe this is a potential future application. For instance, Heirinch et al. used a 3D bioprinted GBM model to study GBM and immunologic cells crosstalk (Heinrich et al., 2019).

Microfluidics integration with bioprinting is an emergent strategy to generate completer and more biomimetic models. For instance, microfluidic chips can be manufactured and cultured through 3D-bioptinting, generating more dynamic, flexible and reproducible systems (Datta et al., 2020).

Other recently reported bioprinted models include GBM cells and stromal cells other than ECs, as macrophages (Grolman et al., 2015; Heinrich et al., 2019). These models allow the communication between different TME cells and the testing of drugs targeting this intercellular communication. Therefore, in the future, it is possible for these models to become more complete, dynamic and closer to the native TME.

Perfusable in vitro models could also be useful to study drug transport and distribution kinetics, after its inclusion in a microcirculatory system. In line with this, platforms have already been designed to mimic drug transport in the bloodstream and passage through vessel walls, followed by crossing of the interstitial fluid and entry into tumor cells (Cao et al., 2019). In the future, it would be interesting if these types of models could be adapted to represent GBM and to assess drug distribution after intravenous administration.

To date, one of the most promising alternatives for in vitro modeling of vascularized tissue is the use of microfluidic systems-based models. Microfluidic systems are devices with networks of micro-channels where liquids can flow in controlled manners. Usually the channels are made of biocompatible materials as polydimethylsiloxane (PDMS), PEG or polystyrene (PS). Their biocompatibility along with other desirable features, such as easy tunability and cost-effectiveness, promoted the association of these platforms with tissue engineering methods to recreate the complex human physiology—the so called organ-on-a-chip technology (Bhatia and Ingber, 2014). Compared with spheroids or non-bioprinted scaffold-based models, microfluidic-based models are less variable in terms of size and shape (Bhatia and Ingber, 2014). Furthermore, in these models, cells are exposed to mechanical signals, such as tension, compression or microvascular blood flow shear stress (1–10 dyne/cm³), thus offering the possibility to overcome the limitations of non-perfusable models (Kim et al., 2013).

Similar to other models, organ-on-a-chip technology has been employed in the establishment of in vitro GBM vascularized models, useful in studies of fundamental biology, such as tumor-associated vascularization processes. Usually, different channels are cultured with tumor cells or ECs, allowing their communication without direct contact. Through paracrine communication, tumor cells induce perfusable microvascular networks development (Table 1) (Kim et al., 2013; Truong et al., 2019). Thus, enabling neovascularization processes study. With the incorporation of patient-derived cells, researchers can study neovascularization processes from an individual perspective (Figure 6) (Table 1) (Amemiya et al., 2021). Furthermore, these models are useful for the study of tumor cells behavior in perivascular niche, namely migration and invasion towards vascular elements (Table 1) (Ayuso et al., 2017; Chonan et al., 2017; Truong et al., 2019).

By co-culturing tumor cells and endothelial cells in distinct channels (Table 1), it was observed tumor cell migration and invasion towards vascular structures. Furthermore, researchers identified the SDF-1 signaling pathway as a major contributor for GSC invasion. Through binding to its receptor, CXCR4, endothelial SDF-1 promoted GSC invasion, whereas GSC treatment with a CXCR4 inhibitor significantly reduced their invasion (Truong et al., 2019). Therefore, microfluidic-based models not only are able to recapitulate the typical architecture of the GBM perivascular niche, but are also suitable for drug screening purposes, especially to test drugs that interfere with endothelial-tumor cell communication.

In addition, microfluidic-based models are promising tools to represent and study GBM hypoxic niches. By blocking the channels that are feeding tumor cells, it was observed the migration of tumor cells towards the opposite direction (Table 1) (Ayuso et al., 2017).

Finally, we would also like to point out that other stromal elements have been included in microfluidic-based GBM models, bringing them closer to the in vivo tumor. Indeed, Cui et al. engineered a sophisticated microfluidic system including not only patient-derived cells and endothelial cells, but also immune cells. Macrophages and tumor cells incorporated a spheroid while T cells were circulating in an adjacent channel (Table 1). This model generated a GBM immunosuppressive heterogeneous microenvironment since T human cells extravasation and infiltration was variable between patients and dependent on tumor-immune-vascular cell crosstalk. Apparently, macrophages present within the tumor acquired an anti-inflammatory phenotype, inhibiting T cell extravasation and infiltration. Nevertheless, T cells were found in most of the tumors, but they were inactive. Cui et al.’s model also helped to monitor real-time immune cell activity and response to several immunotherapies (Cui et al., 2020).

Taking in consideration their applicability, microfluidic-based models present an unquestionable value in biomedical research. They can accelerate the identification of novel biomarkers/therapeutic targets and personalized therapeutic strategies. However, these systems face same limitations too. There is a limited control over their size and over the arrangement of different cell types within the system (Yu et al., 2019). Their use depends on several other equipment, namely pumps and valves (Yu et al., 2019). Furthermore, scale-up is difficult. So, in the future, we believe new and improved microfluidic systems will emerge to overcome these limitations.