Dissecting and rebuilding the glioblastoma microenvironment with engineered materials

Abstract

Glioblastoma (GBM) is the most aggressive and common form of primary brain cancer. Several decades of research have provided great insight into GBM progression; however, the prognosis remains poor with a median patient survival time of ~ 15 months. The tumour microenvironment (TME) of GBM plays a crucial role in mediating tumour progression and thus is being explored as a therapeutic target. Progress in the development of treatments targeting the TME is currently limited by a lack of model systems that can accurately recreate the distinct extracellular matrix composition and anatomic features of the brain, such as the blood-brain barrier and axonal tracts. Biomaterials can be applied to develop synthetic models of the GBM TME to mimic physiological and pathophysiological features of the brain, including cellular and ECM composition, mechanical properties, and topography. In this Review, we summarize key features of the GBM microenvironment and discuss different strategies for the engineering of GBM TME models, including 2D and 3D models featuring chemical and mechanical gradients, interfaces and fluid flow. Finally, we highlight the potential of engineered TME models as platforms for mechanistic discovery and drug screening as well as preclinical testing and precision medicine.

TOC blurb

Glioblastoma is the most aggressive form of brain cancer. This Review surveys the role of biomaterials-based models of the glioblastoma microenvironment, which plays a crucial role in tumour progression, in the advancement of the understanding of the tumour–microenvironment interaction and in the development of effective treatments.

Introduction

Glioblastoma (GBM) is the most common and aggressive primary central nervous system (CNS) tumour, with a devastatingly low median patient survival of 15 months (BOX 1).^1,2 Standard treatment consists of surgical resection followed by chemotherapy and radiotherapy.³ However, GBMs exhibit a diffuse invasion pattern in which tumour cells either migrate individually or collectively infiltrate healthy tissue beyond the tumour margin⁴, making complete surgical resection virtually impossible.⁵ Radiotherapy protocols cover a 2 cm margin beyond the visible tumour margin; however, microscopic tumour invasion may spread beyond this distance.⁶ Infiltrating tumour cells are enriched with glioblastoma stem cells (GSCs), which are tumour cells characterized by their ability to recapitulate the vast heterogeneity of GBM cell phenotypes through propagation and differentiation.⁷ GSCs are often highly refractory to chemotherapy, driving tumour recurrence and chemoresistance.⁸ The tumour microenvironment (TME), which contains extracellular matrix (ECM), interstitial fluid and various stromal cells (for example, astrocytes, macrophages and endothelial cells), is a key regulator of tumour progression.⁹ Substantial advances have already been made in understanding microenvironmental contributions to the progression of other cancers, particularly breast cancer^10–13 and pancreatic cancer.^14,15 Therefore, new therapies have also been developed to target the GBM TME.^{16,17, 269}

BOX 1. Clinical overview of glioblastoma.

Glioblastoma (GBM) comprises 47.7% of all malignant primary central nervous system tumours with a 5 year patient survival of 5.6%.¹ About 95% of patients are diagnosed after age 40 (median age = 65), and no genetic predispositions are known.²⁵⁶ GBM driver mutations can be traced to astrocyte-like neural stem cells in the subventricular zone²⁵⁷; notably, targeting radiotherapy towards the subventricular zone improves patient outcome.^258,259 Primary GBM tumours arise de novo and account for 90% of cases, whereas secondary tumours arise from lower-grade gliomas and account for 10% of cases.²⁶⁰ Secondary tumours are typically diagnosed in younger patients (mean age 45 years) and correlate with longer survival.^1,260 Patients with both primary and secondary tumours typically present symptoms of increased intracranial pressure, such as headaches, neurological defects and seizures.¹⁰⁹ The diagnosis of GBM is based on the presence of several histological features including anaplasia, mitotic activity, microvascular proliferation and necrosis.²⁶¹ Isocitrate dehydrogenase (IDH)-mutant status correlates with secondary GBM and better prognosis, possibly because IDH mutation increases genome-wide methylation.^262,263

Standard treatment is surgical resection followed by chemotherapy and radiation.³ Surgical resection provides clinical relief, enables tissue acquisition for diagnostic analysis and increases survival.⁵ However, complete surgical resection is virtually impossible and must be balanced with preserving intact tissue.²⁶⁴ Since 2005, alkylating-agent temozolomide (TMZ) combined with radiotherapy has become the standard-of-care for newly diagnosed GBM.^3,265 Methylation of the promoter necessary to express O⁶-methylguanine methyltransferase (MGMT), a DNA excision repair enzyme, suppresses reversal of TMZ-induced DNA damage and correlates with increased survival.²⁶⁶ Despite initial efficacy, tumours ultimately acquire therapeutic resistance and recur.⁸ Nitrosureas or a combination of procarbazine, lomustine and vincristine are second-line treatments owing to their higher toxicity and poorer efficacy compared to TMZ.^267,268 Bevacizumab, an antibody-based antiangiogenic therapy, which normalizes the vasculature, was FDA-approved for recurrent GBM in 2009, but was ultimately ineffective at treating GBM in randomized clinical trials.^99,151–153 Steroids, specifically dexamethasone, are prescribed throughout treatment to ameliorate peritumoral edema and discomfort.⁵

Unique features of the brain TME include the blood-brain barrier (BBB), the presence of myelinated and interconnected axon tracts, and a distinct ECM composition, all of which pose specific challenges for treatment.^9,18,19 The BBB, even after losing integrity during tumour progression, is impassable for most chemotherapeutics²⁰ and is especially impermeable in the actively invading tumour regions, where the BBB is intact.²¹ Haptotactic cues from the vascular basement membrane and enrichment of vascular-derived chemotactic cues further drive cell invasion and therapeutic resistance of tumour cells in the perivascular space.¹⁸ Interconnected axon tracts also provide haptotactic cues for cellular invasion and represent a major barrier to surgical resection.^22,23 Furthermore, in contrast to other solid tissues, brain ECM is particularly soft (300 – 3000 kPa^24,25), lacks collagen fibres and is rich in hyaluronic acid (HA), tenascins and chondroitin sulfates.¹⁹ Interestingly, GBMs rarely intravasate and metastasize from the brain, possibly owing to early patient mortality or the unique features of the brain TME.²⁶

Investigations of TME–tumour interactions are limited by a lack of model systems that accurately represent the human brain microenvironment. Biomaterials and engineered devices offer the possibility to recreate brain-like TMEs, enabling mechanistic discovery and therapeutic screening in environments that mimic tissue more closely than traditional 2D culture paradigms. For example, standard tissue culture plastic and reconstituted basement membrane preparations lack design flexibility and fail to capture key compositional, structural and mechanical features of the brain TME.^27–29 Furthermore, engineered TME models can be tailored to incorporate patient-derived cells and matrix, offering a route towards precision medicine. In this Review, we summarize how the TME drives GBM progression, describe potential therapeutic targets and investigate designs and applications of engineered TME models in research and the clinic. Finally, we outline new directions for designing, fabricating and employing engineered models in patient care.

Glioblastoma microenvironment

The TME provides a dynamic array of signals that drive proliferation, invasion and resistance (FIG. 1). These signals can be broadly categorized into ECM composition, ECM mechanics, topographical cues, interstitial fluid and stromal-cell interactions (TABLE 1).

Figure 1. Schematic of gliobastoma (GBM) regions.

This GBM schematic illustrates changes during tumour progression in the different microenvironmental regions. a) The necrotic core is softer than surrounding tissue and is thought to form after increases in cell density beyond a certain threshold or vaso-occlusive events result in hypoxia. b) Psuedopalisades are regions of high cell density thought to form as cells migrate away from hypoxic regions. These zones have an increased elastic modulus and matrix production compared to healthy tissue and necrotic regions. GBM cells invade from the outer edge of the cell-dense tumour into healthy tissue at the infiltrating rim. c) GBM tumours show hypervascularity with increased angiogenesis compared to healthy brain tissue. Tumour-associated vasculature is poorly formed, leaky and leads to an increase in interstitial fluid pressure. d) Tumour cells invading through the parenchyma often follow and remodel the surface of myelinated tracts – a region in which high interstitial fluid flow may also drive invasion. e) Tumour cells rapidly invade the vasculature, where they are exposed to nutrients, high interstitial fluid flow and haptotactic cues in basement membranes. The perivascular niche also supports stemness and survival of glioblastoma stem cells (GSCs). TAM, tumour-associated macrophage; ECM, extracellular matrix; BBB, blood-brain barrier.

TABLE 1.

Key signals in the tumour microenvironment.

Signal type	Signal	Signalling effects	Effect on tumour progression	Refs.
Matrix composition	HA	GBM cells increase HA synthesis and degradation	Low MW HA accumulates and promotes GBM cell invasion, GSC stemness and GSC resistance	31,37–45
	Fibronectin	GBM cells decrease fibronectin expression and crosslinking	Invasion and sensitivity to therapy increase	51–54
	Tenascin C	GBM cells express more tenascin C	Tenascin C increases matrix stiffness and GBM cell invasion and proliferation	60–62
	Laminin	GSCs interact with laminin	GSCs show increased stemness, invasion and proliferation	48–51
Matrix mechanics	Elastic modulus	Elastic modulus increases in pseudopalisades and decreases in necrotic core compared to healthy tissue	Increased modulus promotes GBM cell migration and proliferation in vitro	69–75,77,78
	Density	GBM cells produce more matrix than non-tumour cells	High matrix density decreases perfusion and increases ECM compaction and cell damage	79,80
Topography	Microvasculature	Tumours exhibit hypervascularity with loss of BBB integrity and change in basement membrane composition	Tumour cells invade rapidly along vasculature	63,64
	Myelinated tracts	GBM cells remodel myelin coating	GBM cells invade rapidly along myelinated tracts	84–88
Interstitial fluid	Pressure	Tumours exhibit edema	Pressure from edema is a barrier to chemotherapy	91,92
	Fluid flow	Convection-enhanced therapy increases flow rates	Fluid flow promotes invasion and proliferation	93–97,170
Stromal and endothelial cell crosstalk	TAMs	GBM-derived osteopontin recruits and maintains TAM phenotype; TAMs secrete complex array of cytokines and growth factors	Immune activity (from cytotoxic T cells) increases; growth factors increase GBM proliferation, survival and migration	103,104
	TAAs	GBM cells activate TAAs; TAAs activate tumour cell MMP and uPA expression	Intratumoural immune response decreases; GBM invasion increases and cells become more chemoresistant	102
	Vascular endothelial cells	Vascular endothelial cells secrete IL-8	GSC migration, proliferation and stemness increase	106
	Neurons	Neurons secrete neuroligin-3	GBM proliferation increases	105
	MSCs	MSCs provide exosome cargo such as miR-1587 and secrete IL-6	GSCs proliferation and tumour cell survival increase	100,101

GBM, glioblastoma; GSC, glioblastoma stem cell; HA, hyaluronic acid; IL, interleukin; miR, microRNA; MMP, matrix metalloprotease; MSC, mesenchymal stem cell; MW, molecular weight; TAA, tumour-associated astrocyte; TAM, tumour-associated macrophage, TME, tumour microenvironment; uPA, urokinase-type plasminogen activator; ECM, extracellular matrix.

Extracellular matrix

Normal brain ECM, in contrast to the ECM of other solid tissues, is enriched in glycoproteins, such as tenascins and link proteins, glycosaminoglycans (GAGs), such as HA, and proteoglycans, such as aggrecan, neurocan, versican and phosphacan.³⁰ Conversely, fibrillar proteins, such as collagen and fibronectin, are relatively sparse.³¹ In tumours, the abundance of ECM components is altered; in particular, the level of GAGs is increased by 3–4 fold.³² Astrocytes and oligodendrocytes produce the majority of brain ECM in normal tissue, but GBM cells also express their own pro-invasive matrix.^18,33 GBM cells can also induce stromal cells to express specific ECM components. In highly angiogenic tumours, tumour cells overexpress tenascins and vitronectin, and stromal cells produce excess laminin, fibronectin and collagen IV.³⁴

HA, a polyanionic GAG localized primarily in the intraparenchymal region, is the most abundant component of brain ECM.³¹ Expressed as a megadalton linear chain in healthy tissue, HA regulates tissue mechanics, organization and hydration. HA also activates cellular signalling through surface receptors such as CD44 and receptor for hyaluronan mediated motility (RHAMM).^35,36 The differential signal transduction and functional contributions of CD44 and RHAMM remain incompletely understood; however, it is known that both receptors can drive invasion.^37–39 Both tumour and stromal cells produce HA in high-grade gliomas, and GBMs overexpress hyaluronan synthase 2 (HAS2).^40–42 Whether downstream signals arising from HA–receptor interactions are pathologic is determined by the molecular weight of HA; low-molecular weight HA provides pro-invasive cues and high-molecular weight HA reduces tumour invasion.^43,44 Accordingly, GBM spheroids are less invasive in 3D matrices crosslinked with 500 kDa HA than with 60 kDa or 10 kDa HA.⁴⁵ The crucial role of HA in GBM progression motivates the investigation of the effects of the molecular weight, mechanical properties and signalling of HA in engineered TME models.

Laminin, fibronectin and collagen IV are mainly localized in vascular basement membranes.^19,46 Laminin has been shown to be particularly potent in driving GBM progression; however, downstream signalling mechanisms may be isoform-specific.⁴⁷ For example, in a zebrafish model, laminin α5 increases the formation of blood vessel-dependent tumours, but reduces the migration speed of GBM cells⁴⁸. In human cell culture models, laminin α2 supports GSC growth.⁴⁹ Interestingly, GSCs are often propagated on laminin-coated culture dishes, and laminin-binding integrin α6 is necessary for GSC renewal, proliferation and tumour formation.⁵⁰ By contrast, fibronectin expression is often decreased in GBMs.⁵¹ Fibronectin assembly reduces GBM cell migration and fibronectin depletion increases migration.^52,53 Pharmacological disruption of fibronectin assembly in orthotopic mouse models also sensitizes tumours to chemotherapy.⁵⁴ Thus, assembled fibronectin may inhibit GBM cell invasion but may also reduce the efficacy of chemotherapy. Whether targeted disruption of fibronectin would advance or counteract therapeutic goals remains unclear. Fibrillar collagens, such as collagen I, are not abundant in normal brain tissue; however, non-fibrillar collagen IV is present in basement membranes of the brain vasculature.^55,56 Despite widespread use in engineered TME models^27,29, the role of parenchymal collagen in GBMs in vivo is unclear. Evidence suggests that the structural organization of collagen has an influence on GBMs; accumulation of punctate or non-fibrillar collagen can be correlated with a more invasive phenotype than accumulation of organized fibrillar collagen, which may structurally impede parenchymal invasion.⁵⁷

The brain also contains matricellular proteins, which regulate tissue structure and tumour invasion.³⁰ Tenascin C, which is a large (180–250 kDa) glycoprotein that crosslinks matrix, is particularly important in GBM progression.^58,59 Aggressive gliomas are enriched in tenascin C, which correlates with poorer patient prognosis.⁶⁰ Interestingly, glioma ECM stiffness also corresponds with levels of tenascin C, but not with levels of type I collagen abundance, vascularity or tumor cell density.⁶⁰ Tenascin C further participates in cell–cell crosstalk. Tumour cell-derived tenascin-C interacts with α5β1 and αvβ6 integrins on T lymphocytes, resulting in reduced mTOR signalling and immunosuppression.⁶¹ Additionally, the presence of tenascin C in collagen I matrices leads to an increase in matrix metalloproteinase (MMP)-12-mediated GBM invasion.⁶² Other matricellular proteins, notably agrin, insulin-like growth factor-binding protein (IGFBP) 7 and secreted protein acidic and rich in cysteine (SPARC), are dysregulated in GBM vascular basement membranes, which may contribute to the disruption of the BBB and angiogenesis.^63,64 The matricellular protein osteopontin (Spp1) is further implicated in promoting GBM therapeutic resistance. Osteopontin affects the permissiveness of the TME, and maintains the stemness of GSCs through CD44-dependent signalling in the perivascular space.^65–67

The expression of these different ECM components is highly intertwined. For example, silencing uridine diphosphate glucose 6-dehydrogenase (UGDH), which is an enzyme required for GAG monomer synthesis, results in decreased GAG production and abundance of tenascin-C and brevican, leading to a reduction of tumour growth and migration in animal models.⁶⁸ Therefore, dissecting the complexity of matrix composition in engineered TME models may uncover targetable drivers of GBM progression.

The mechanical properties of the tumour ECM, for example matrix density and bulk storage modulus, also play an important role in GBM progression. Like most tumours, GBMs also exhibit an elastic modulus almost twice that of normal tissue, possibly owing to changes in ECM expression and increased compaction.^69,70 However, the elastic modulus varies strongly by region, with a lower modulus observed in necrotic regions (~0.1 kPa) than in the hypercellular core (~10 kPa) and a higher modulus observed in the hypercellular core compared to healthy tissue (1 kPa).⁷¹ Notably, GBM cell proliferation and migration is mechanosensitive^72,73, although the degree of mechanosensitivity varies between patients.⁷⁴ The mechanosensitivity also differs between tumour cell subpopulations, and some GSCs lack mechanosensitivity.^75,76 High matrix modulus (6.9 kPa compared to 0.15 kPa) induces CD44-dependent cell migration and spreading on HA.⁷⁷ High matrix modulus (119 kPa compared to 0.08 kPa) also amplifies epidermal growth factor receptor (EGFR) signalling, promoting proliferation.⁷⁸ Matrix density is also higher in GBMs than in healthy brain tissue, perhaps owing to compaction caused by matrix overexpression and high cell density. Compaction of GBM cells in vitro further induces expression of collagen IV and VI, vascular endothelial growth factor (VEGF) and the collagen-crosslinking enzyme lysyl oxidase, which is associated with an increase in angiogenesis and matrix elastic modulus.⁷⁹ The growing tumour mechanically compresses tissue, damaging neurons and restricting vascular perfusion.⁸⁰ GBM ECM remodelling progresses as a positive feedback loop in which tumour cell proliferation and ECM production cause an increase in elastic modulus, which in turn further promotes tumour cell proliferation and invasion.

Tumour–stroma interactions

GBM cells most rapidly invade along anatomical tracks, such as the vasculature and myelinated axons (FIG. 1c, d & e).^19,23 As GBM cells invade through the perivascular space along the vascular basement membrane, they disrupt astrocytic end feet contacts with endothelial cells and weaken the BBB (FIG. 1e).⁸¹ A combination of haptotactic, chemotactic and topographic cues are likely responsible for this pattern of invasion. Many integrin-binding matrix proteins, such as laminin, collagen and fibronectin, are localized at the vascular basement membrane and are relatively sparse in other brain regions.^19,46 Basement membranes have a higher elastic modulus than the surrounding matrix, which may promote a mechanosensitive, integrin-mediated migration.⁸² The perivascular space is also rich in paracrine signals from perivascular support cells as well as nutrients crossing the BBB.⁸³ The detailed mechanisms of invasion along myelinated axon tracts remain elusive thus far; however, MMP-mediated remodelling of myelin from a non-adhesive to an adhesive substrate is likely involved.^84–87 GSCs that migrate along remodelled or deteriorating white matter tracts gain access to the Notch ligand Jagged1 on exposed nerve fibres, which further promotes invasive growth.⁸⁸ Culturing GBM cells on engineered surfaces with linear topographies shows that linear presentation of ECM cues strongly affects migration speed. The resulting constraint and alignment of actin bundles as well as cytoskeletal polymerization coordinate rapid, persistent migration.^89,90

Solid tumours exhibit an abnormally high interstitial fluid pressure and volume mainly owing to leaky vasculature.^91,92 Interstitial fluid flow is most rapid along axon tracts and in perivascular spaces, promoting the distribution of soluble cues, for example, pro-angiogenic factors.⁹³ Rapid flow in parallel with white matter tracts leads to an increase in the invasion speed of tumour cells, possibly owing to shear stress or to effects on soluble cue gradients.⁹⁴ In vitro and in vivo studies show that interstitial fluid flow promotes migration mediated by the CXCR4 receptor and, to a lesser degree, by CD44–HA interactions.^95–97 The composition of interstitial fluid substantially varies by tumour region. Lack of dissolved oxygen (hypoxia) and low pH are characteristic of interstitial fluid in the tumour core, which perpetuates necrosis and drives tumour cells towards invasive and pro-angiogenic phenotypes.⁹⁸ The high interstitial fluid pressure (IFP) in solid tumours is further a major barrier to chemotherapeutic delivery, because it prevents the transport of small molecules into the tumour core.⁹¹ Some therapeutic treatments cause a decrease in IFP, which could improve the therapeutic efficacy and reduce edema. In particular, treatment with Bevacizumab in orthotopic GBM models causes a reduction in IFP by ~73%, likely owing to a normalization of the vascularity (BOX 1).⁹⁹ The importance of interstitial fluid in GBM is well-established; however, therapeutic interventions to target interstitial fluid are limited.

Tumour cells and stromal cells in the TME co-evolve during tumour progression. Immune and inflammatory cells, such as infiltrating monocytes and fibroblasts, endothelial cells and glioma-associated mesenchymal stem cells (MSCs), which are located throughout the tumour and in the intraparenchymal region, interact with tumour cells, driving disease progression (FIG. 1). Tumour cells also interact with other intraparenchymal stromal cells, such as astrocytes, pericytes, oligodendrocytes and neurons. A common and crucial function of these non-tumour cells is to secrete signals that modulate tumour cell survival, proliferation and migration. For example, MSCs secrete exosomes and soluble cytokines, such as interleukin-6, which interact with GSCs, increasing their proliferation and stemness.^100,101 Tumour-associated astrocytes (TAAs) release secreted factors that support tumour cell survival and proliferation, modulate the intratumoural immune response and promote invasion by activating tumour-derived matrix-remodelling enzymes, including MMPs and urokinase-type plasminogen activator (uPA).¹⁰² GBM cells also extensively interact with microglia and infiltrating tumour-associated macrophages (TAMs) to suppress an anti-tumour immune response.^66,103,104 Neurons promote proliferation of GBM cells through secretion of soluble factors such as neuroligin-3.¹⁰⁵ Tumour cells also closely interact with vascular endothelial cells (FIG. 1c,e). For example, endothelial cells secrete interleukin-8 and GSCs upregulate interleukin-8 receptors, which stimulates migration, growth and stemness.¹⁰⁶ Tumour cells can further directly participate in vessel mimicry by aligning with endothelial cells to form vascular walls or by transdifferentiating into endothelial cells.^107,108 Therefore, the incorporation of the stromal secretome in engineered TMEs is important owing to its crucial role in regulating tumour cell behaviour, particularly in the context of immunotherapy.

Targeting the microenvironment

The TME substantially changes over time and in the different microregions, particularly during therapeutic treatment. Magnetic resonance imaging (MRI) scans of newly diagnosed patients typically reveal a contrast-enhancing, irregularly shaped GBM tumour border with pseudopalisades, or regions of high cell density, surrounding a hypointense region of necrosis (FIG. 1a&b).¹⁰⁹ Necrotic cores are thought to arise once the tumour cell density exceeds a certain threshold at which the cells cannot be anymore supported by diffusion-based transport of nutrients, gases and metabolites from deteriorating or occluded vasculature. As cells migrate away from hypoxic regions, pseudopalisades form and recruit new vasculature (FIG. 1b&c).¹¹⁰ As the tumour grows and invades, the adjacent tissue deteriorates (FIG. 1d&e). Neurodegeneration is caused not only by mechanical stresses⁸⁰ but also by aberrant levels of tumour-secreted soluble factors, such as the extracellular domain of CD44.¹¹¹ Surgical resection of >98% of the gross tumour, including necrotic and pseudopalisading regions, increases overall patient survival.¹¹² Metabolic, fluorescent dyes can be employed during surgery to improve the identification of the tumour edge, although the clinical benefit is not yet clear.¹¹³ Carmustine-releasing Gliadel wafers can be implanted following surgical resection and may especially benefit patients for whom gross resection is unfeasible; however, the efficacy and safety of this approach remain controversial.^114,115 Tumour-treating fields (alternating electric fields) that disrupt mitosis may also improve patient survival.^116,117

Glioblastoma stem cell niches

The resection of diffusely invading cells beyond the gross tumour edge poses risks of destroying functional tissue. Even if resection is performed beyond the tumour edge, there is no assurance that all tumour cells can be located and resected.⁵ The clinical need for therapies targeting the remaining tumour cell population has motivated the investigation of how the TME promotes survival, invasion and proliferation of diffusely infiltrating tumour cells. GSCs are especially adept at invading healthy tissue and resisting chemo- and radiotherapy, which makes them a key candidate for targeted adjuvant therapies. GSCs reside within specific anatomic niches, which are specialized microenvironments that regulate GSC stemness, proliferation and apoptosis resistance, analogous to tissue stem cell niches.^83,118–120 Importantly, these niches shield GSCs from anticancer therapies by providing pro-survival cues and by anatomically blocking them from therapy exposure.¹²¹ Four unique zones (subarachnoid, perineuronal, perivascular and perinecrotic) have thus far been identified that support GSC self-renewal and proliferation.¹²⁰ Each zone has a distinct TME composition with niche-specific transcriptional and epigenetic signatures.^119,120

The contributions of the perivascular niche to therapy resistance, infiltration spread and disease progression are perhaps best understood.^{83,118,122–124} In the perivascular niche, GSCs and the TME engage in cooperative signalling, promoting neovascularization and GSC maintenance. The leaky vasculature provides access to nutrients, and the endothelium activates Notch-dependent pathways that promote GSC self-renewal and therapy resistance.¹²⁵ In turn, GSCs support neovascularization by secreting angiogenic factors such as VEGF.¹²⁶ Interestingly, endothelial-derived nitric oxide increases the tumour-initiating capacity of the platelet-derived growth factor receptor (PDGFR)-expressing subset of GSCs.¹²⁷ Matrix composition and mechanics of the perivascular niche also drive GSC tumorgenicity.⁸³ In particular, HA regulates GSC stemness by engaging the HA-specific cell receptor for hyaluronan-mediated motility (RHAMM)¹²⁸ and CD44 and by activating the transcription of stemness modulators.¹²⁹ HA also activates the toll-like receptor (TLR) 4–nuclear factor (NF)κB pathway to promote stemness; the expression of TLR4 receptors is upregulated during GSC differentiation along with HA synthesis, which increases NFκB activity and suppresses terminal GSC differentiation.¹³⁰ Furthermore, altered mechanotransduction caused by niche remodelling stimulates GSC tumorgenicity.¹³¹ For example, a pro-tumorigenic glycocalyx-integrin feedforward loop, in which ECM stiffening induces a mesenchymal transition in GSCs, drives GBM progression correlated with poor prognosis.^132–135 In a brain-mimetic biomaterial platform for the 3D culturing of patient-derived GBM cells, the modulation of both the HA content and of the mechanical properties of the biomaterial are required to recreate the known resistance of GBM cells to the EGFR inhibitor erlotinib, highlighting that the TME can diminish therapeutic efficacy.¹³⁶

Although less understood, hypoxic GSC niches also substantially contribute to the maintenance of GSC populations.^98,137,138 Hypoxic niches arise when defective vessels are obstructed or collapse, which leads to a reduction in oxygenation.¹³⁸ Cells adapt to low oxygenation by activating hypoxia inducible factors (HIFs).⁹⁸ Activation of HIF-1α promotes GSC self-renewal and growth and causes pro-invasive protein expression through upregulation of C-X-C chemokine receptor type 4 (CXCR4), which is a chemokine receptor related to increased migration.¹³⁷ Similarly, HIF-2α promotes the expression of Oct4, which is a stem cell marker strongly associated with stemness.¹³⁹ Interestingly, HIF-2α is specifically expressed by GSCs and thus may serve as a potential GSC-specific marker.¹³⁹ Hypoxia may even promote the reprogramming of non-stem GBM cells towards a GSC-like phenotype.¹³⁹ Therefore, TME niches play a multifaceted role in regulating GSCs, motivating their investigation in engineered TME models.

Microenvironmental changes

Radiotherapy increases overall patient survival by reducing tumour burden and by improving BBB permeability for chemotherapeutics; however, radiotherapy also triggers the remodelling of the TME, which increases the aggressiveness of tumours at recurrence.¹⁴⁰ In response to radiation, TAMs infiltrate the tumour through the defective BBB and astrocytes adopt a reactive phenotype, which induces tissue inflammation.¹⁴⁰ Moreover, in contrast to bulk tumour cells, GSCs are particularly efficient at evading radiotherapy by activating DNA damage checkpoints to repair DNA damage.¹⁴¹ The TME promotes tumour cell survival during radiation treatment; for example, in a co-culture of GSCs with astrocytes, signal transducer and activator of transcription 3 (STAT3) signalling is activated in GSCs in response to astrocyte-secreted factors, which increases GSC radiation resistance.¹⁴² Radiation further temporarily induces senescence in GBM cells by triggering a ‘senescence-associated secretory phenotype’, which leads to upregulation of ECM expression, proteolytic enzymes and pro-inflammatory signalling molecules.¹⁴⁰ After exiting senescence, these cells and their microenvironments are primed for invasion and proliferation. GBM cells increase HA production in response to radiation by increasing the expression of hyaluronan synthase 2 (HAS2), which correlates with increased invasion.⁴¹ Senescence also occurs in stromal cells¹⁴⁰, and tumour cells can compensate for endothelial cell senescence by trans-differentiating into endothelial cells enabling angiogenesis.¹⁴³

The chemotherapeutic temozolomide (TMZ) increases patient survival but can trigger TME remodelling that promotes a resistant, pro-invasive tumour phenotype. Treatment of cultured GBM cells with radiation and TMZ induces an increase in MMP-2 secretion and abundance of matrix-degrading invadopodia.¹⁴⁴ TMZ treatment also alters proteoglycan and GAG composition, with the combination of TMZ and dexamethasone resulting in deterioration of proteoglycan and GAG content.¹⁴⁵ Other agents promote TME remodelling that slows tumour progression. Microtubule inhibitors target cell division, but they can also reduce the invasive capacity of tumour cells by reducing MMP-2 expression.¹⁴⁶ Dexamethasone, which is a steroid traditionally applied for its ability to reduce edema rather than for its chemotherapeutic properties, also activates fibronectin matrix assembly, resulting in increased cell–cell and cell–matrix adhesions that may slow invasion.⁵¹ However, the role of dexamethasone and other steroids in tumour progression and their interactions with therapeutic interventions are largely unknown. The investigation of treatment-induced TME remodelling in engineered models could unravel these interactions to improve therapeutic strategies.

Targeted therapeutic agents

Targeting therapeutics to the tumour and the TME offer promise to improve patient survival and quality of life.^147,148 Successful clinical treatment of chronic myeloid leukemia and gastrointestinal stromal tumours with the small molecule inhibitor imatinib mesylate (Gleevec) targeting mutated kinases demonstrated the potential of targeted therapies.¹⁴⁷ Targeted therapies have also been clinically successful in breast cancer treatment, particularly for the human epidermal growth factor receptor 2 (HER2)-amplified subset.¹⁴⁸ Unfortunately, most of the clinically tested GBM-targeted therapies have shown little efficacy thus far, such as erlotinib targeting the often overexpressed EGFR or PLX3397 targeting colony stimulating factor 1 receptor (CSF1R) to modulate TAM activity.^16,149,150 Inhibitors targeting the hypervascularity of GBM tumours have come closest to realization and remain a promising strategy (TABLE 2). The anti-VEGFA therapeutic Bevacizumab is currently the only FDA-approved drug targeting the GBM TME.^151–153 Bevacizumab treatment initially causes a decrease in tumour volume and vascularity, but tumours ultimately adapt with revascularization and increased invasiveness.¹⁵⁴ A more potent pan-VEGF family inhibitor, tivozanib, reduces proliferation and invasion and is currently undergoing clinical evaluation.¹⁵⁵ Similarly, inhibitors of VEGF receptor tyrosine kinases, such as Cediranib and Sunitinib, show promise in reducing angiogenesis and normalizing vascularization.^156–158 Other angiogenic targets are also under investigation; for example, the angiopoietin inhibitor AMG 386 reduces vascular permeability and angiogenesis.^159,160 The potential of antiangiogenic therapies motivates the investigation of vascular–tumour interactions in engineered TME models.

TABLE 2.

Selected therapeutics targeting the tumour microenvironment in clinical trials.

Therapeutic agent	Target	Effect on tumour progression in preclinical models	Refs.
Microglia and TAMs
PLX3397	CSF1R inhibitor	↓Microglia, ↓tumour burden, ↓invasion	150
Cell receptor-ECM interactions
Cilengitide	Pentapeptide that blocks activation of αvβ3 and αvβ5 integrins	↓Angiogenesis and tumour growth by blocking of integrins on vascular endothelial and tumour cells	167
Hypoxia
AQ4N	Bioreductive prodrug targeting topoisomerase II in hypoxic cells	↓Hypoxic cells	163
Microvascular-related pathways
Tivozanib	Pan-VEGFR tyrosine kinase inhibitor	↓Proliferation, ↓expression of VCAM-1 and ICAM-1 mediated cell-cell adhesion, and ↓MMP-2-mediated invasion	155
Sunitinib	PDGFR and VEGFR inhibitor	↓Angiogenesis, ↓proliferation	158
Cediranib	Pan-VEGFR tyrosine kinase inhibitor	↓Angiogenesis, normalization of vasculature	156,157
AMG 386	Angiopoietin-1/−2-neutralizing peptibody	↓Vessel permeability, ↓angiogenesis	159,160

CSF1R, colony stimulating factor 1 receptor; ICAM-1, intercellular adhesion molecule 1; MMP-2, matrix metalloproteinase 2; PDGFR, platelet-derived growth factor receptor; TAM, tumour-associated macrophage; TME, tumour microenvironment; VCAM-1, vascular cell adhesion molecule 1; VEGFR, vascular endothelial growth factor receptor; ECM, extracellular matrix.

Several other TME features are also explored as targets (TABLE 2). Efforts to eradicate hypoxic cells within the TME have overall been positive in clinical trials in patients with advanced solid tumors.^161–164 Bioreductive prodrugs can be enzymatically reduced in hypoxic regions into cytotoxic products. AQ4N is a bioreductive prodrug targeting topoisomerase II, and it has shown promise as an adjuvant therapy in preclinical trials of several cancers, including GBM.¹⁶⁴ Importantly, AQ4N can cross the BBB and was well tolerated in all patients in a phase I study in GBM.¹⁶³ Cell–matrix interactions represent another key target for therapies.^165–167 Cilengitide is the first integrin inhibitor undergoing clinical testing and initially showed promise for modestly improving survival in both newly diagnosed and recurrent GBM with tolerable toxicity.¹⁶⁷ Cilengitide inhibits integrins αvβ3 and αvβ5, which are overexpressed on GBM cells and vascular endothelial cells. This inhibition disrupts angiogenesis and tumour–matrix interactions needed for migration. However, Cilengitide was eventually shown to be ineffective in phase III clinical trials¹⁶⁸, which may be related to poor bioavailability; thus, Cilengitide may warrant further investigation.¹⁶⁹ Careful consideration of how the TME influences tumour mechanics and transport can be leveraged to improve drug delivery methods.¹⁶⁵ For example, convection-enhanced delivery involves catheter insertion directly into the tumour core to continuously deliver a chemotherapy, avoiding perfusion across the BBB and counteracting resistance from increased interstitial pressure.¹⁷⁰ Moreover, a poliovirus-based immunotherapy designed to activate oncolytic T-cells has shown promise in improving GBM patient survival and may be combined with molecularly-targeted therapeutic stragies.^171,172

Engineering microenvironment models

Experimental models for GBMs range in complexity from 2D cultures on glass or plastic to orthotopic xenografts and genetically-engineered mouse models.²⁹ Traditional 2D models have proven invaluable for investigating some molecular mechanisms governing GBM progression, such as early studies elucidating how MMPs and soluble factors contribute to tumour initiation, invasion and propagation.¹⁷³ However, 2D models lack the ECM stiffness and composition, topographical guidance cues and dimensionality of human tissue needed to fully investigate the role of the TME. Orthotopic xenografts of patient-derived GBM cell lines in immunodeficient murine models are commonly used to fully recapitulate the in vivo TME. Orthotopic xenograft models better mimic tumour heterogeneity than in vitro models, with different levels of tumour heterogeneity depending on the model.^174,175 However, orthotopic xenograft models lack a normal immune response, which is a key parameter in regulating tumour progression and full retention of tumour heterogeneity.^28,176 Furthermore, animal models are more expensive and less scalable than in vitro models, and are often impractical for detailed mechanistic dissection of human pathobiology.¹⁷⁷ The GBM TME substantially affects tumour progression, and thus, engineered TME models offer a valid alternative as experimental GBM models with the potential to overcome the limitations related to animal models.¹⁷⁸ Specific parameters (ECM composition, mechanics, topography and stromal cells) can be incorporated into engineered models to recreate the GBM TME for more precise hypothesis testing (TABLE 3).

TABLE 3.

Engineered glioblastoma models.

Model	Key findings	Refs.
2D matrix models
PA	Spreading, migration and proliferation increases with matrix stiffness, depending on tumour cell subpopulation and patient	72,74,75,78,179
Silicone rubber	Spreading increases with elastic modulus	73
Collagen	Matrix biophysical properties affect phenotype	180,181
HA	CD44 is mechanosensitive; elastic modulus affects miRNA expression	77,182,183
3D matrix models
Collagen	Dimensionality determines drug resistance; Porosity and density affect invasion speed	181,186 189,190
Collagen-agarose	Cell spreading and motility in collagen requires local matrix stiffening	187,188
HA	Cell invasion through HA mimics invasion in the brain and is slow relative to invasion in highly porous matrices	123,183
Matrigel	Stromal cells in 3D matrix affect GBM phenotype	119,167,16
PEG	MMP degradability enhances cell spreading	193
PNJ	Scaffolds increase stemness of GSCs	196
PCL-HA	HA maintains stemness of GSCs	195
Alginate-chitosan	Scaffolds increase stemness marker expression	197
HA-collagen	HA upregulates invasion	207
HA-gelatin	HA upregulates matrix remodelling	45,208
HA-PEG	Matrix elastic modulus affects ECM deposition	209
Brain-derived ECM	Cells exhibit brain-like invasion in matrix	198,199
Models of heterogeneity
Elastic modulus patterning	Higher modulus increases cell spreading in 2D and 3D	215,217
Orthogonal parameter patterning	Composition and stiffness have non-linear effects on phenotype	182,216
Soluble cue gradient	Reduced nutrient and oxygen transport increases secretion of angiogenic factors	218
Topographical models
ECM interface	Interface properties drive invasive morphology	219,220
Open channels	Stiffness and pore size have combined effect on invasion	179
Electrospun fibres	Linear topographic cues drive rapid invasion	194,221–226
Encapsulated fibres or channels	Cells transition to rapid invasion when encountering linear topographic cues in 3D matrix	123,207
Interstitial fluid models
Flow in Boyden chamber	Interstitial flow drives CXCR4-dependent invasion	95–97
Multi-parameter microfluidic system
Pseudopalisade model	Vaso-occlusion drives migration and psuedopalisade formation	228
PVN models	Stromal-cell crosstalk affects invasive phenotype	199,229–231
1. Mini-brain with macrophages	2. GBM cells recruit and influence macrophage polarization	3.238
Organoid
Tumour organoid culture	Tumour organoids maintain heterogeneity and hypoxic gradient	239
Stem-cell derived tissue	Engineered neural tissue supports brain-like GBM invasion	241

CXCR4, C-X-C chemokine receptor type 4; ECM, extracellular matrix; GBM, glioblastoma; HA, hyaluronic acid; miRNA, microRNA; PA, polyacrylamide; PCL, polycaprolactone; PEG, polyethylene glycol; PNJ, poly(N-isopropylacrylamide-co-Jeffamine M-1000® acrylamide); PVN, perivascular niche; GSC, glioblastoma stem cell.

2D matrix models

A simple approach to incorporating TME components into engineered models is to fabricate 2D substrates featuring ECM ligands and mechanical properties normally present in brain matrix. These modified 2D substrates can be used to explore how matrix mechanics and ECM components affect cell morphology, proliferation and migration (FIG. 2a). The mechanical properties of synthetic substrates, such as polyacrylamide^{72,74,75,78,179} and silicone rubber⁷³, can be well controlled in a physiologically relevant range and coated or conjugated with cell-adhesive matrix proteins, such as laminin or fibronectin. Natural or semi-synthetic polymer matrices, such as collagen^180,181 and HA^77,182,183, typically contain some adhesive ECM cues, but they can also be further modified with ligands. HA gels are particularly advantageous for recapitulating the HA-richness of brain ECM. A diverse array of chemistries can be applied in HA gels, such as the addition of methacrylate or thiol groups, to facilitate crosslinking and modification with peptides.^183–185 Synthetic and natural 2D substrates have been applied to demonstrate that GBM cells are mechanoresponsive and that the mechanical response varies between patients and between subpopulations of cells.^74,75 For example, our laboratory has employed 2D HA hydrogels to show that CD44 can transduce mechanical signals from HA to regulate GBM adhesion and invasion.⁷⁷

Figure 2. Engineered glioblastoma models.

a) 2D models often include a matrix layer with tunable mechanical properties and composition. b) In 3D matrices, cells can be encapsulated as spheroids or as single cells. c) Cells can be cultured between extracellular matrix (ECM) layers of distinct composition and mechanics to model cell migration at the interface of the vascular basement membrane and the intraparenchymal matrix. d) Nanofibres with ECM coatings are often used to mimic linear, white matter tracts. e) Media height in a Boyden chamber can be used to generate interstitial flow through matrix-encapsulated cells. f) A microfluidic device with an open (nutrient-rich) and closed (occluded) channel surrounding matrix-encapsulated cells can be used to test how psuedopalisades form. g) A microfluidic model of the perivascular niche (PVN) containing a glioblastoma stem cell (GSC)-rich tumour reservoir, an intraparenchymal region with stromal matrix and a region of matrix-encapsulated endothelial networks can be used to investigate the role of the PVN in GSC tumourigenicity. h) A bioprinted microfluidic model with a matrix-encapsulated endothelial network arranged concentrically around patient-derived tumour cells can be applied for the development of patient-specific engineered tumour microenvironments (TMEs). Panel f adapted from REF. 223. Panel g adapted from REF. 224. Panel h adapted from REF. 195.

3D matrix models

2D platforms can be rapidly fabricated, are parallelizable and amenable to imaging and culture manipulations; however, owing to their 2D nature, they cannot fully capture brain architecture. By contrast, 3D matrices offer the possibility to incorporate soluble cue gradients, such as an oxygen gradient, and confinement of invading cells, which alters cell morphology and requires the cells to degrade or squeeze through the matrix – as is the case in an in vivo TME. Interestingly, dimensionality alone can profoundly affect cell responses to chemotherapeutics, independent of matrix stiffness or composition.¹⁸¹ Materials used for 2D substrates, such as collagen^{181,186–190} and HA^123,183, can also be employed as 3D scaffolds. However, materials such as polyacrylamide or polycaprolactone (PCL) requiring harsh solvents or crosslinking reagents during gelation cannot be easily seeded with cells unless they are made highly porous such that cells can be incorporated into the matrix after gelation. Matrigel, which is a reconstituted basement membrane harvested from mouse sarcoma, is commonly used as 3D matrix because of its rapid, temperature-based gelation, abundance of adhesive sites and compositional complexity.^118,191,192 Collagen and Matrigel are simple to use relative to materials requiring complex synthesis, compatible with 3D cell encapsulation and contain various adhesive sites; however, the collagen-rich composition of both matrices and the fibrous architecture of collagen do not resemble the HA-rich, nanoporous brain matrix. Additionally, Matrigel composition is poorly defined chemically and exhibits batch-to-batch variability. Alternatively, synthetic polyethylene glycol (PEG) gels can be decorated with adhesive peptides and crosslinked with cleavable linkers, enabling precise control over matrix mechanics and composition for GBM modelling. Incorporation of degradability into 3D PEG matrices is not required for GBM cell viability and colony expansion, but is essential for mesenchymal-like cell spreading.¹⁹³ 3D scaffolds, including electrospun polystyrene coated with laminin¹⁹⁴, porous PCL scaffolds with incorporated HA¹⁹⁵, poly(N-isopropylacrylamide-co-Jeffamine M-1000® acrylamide) (PNJ) copolymer scaffolds¹⁹⁶ and electrolyte complexes of alginate and chitosan,¹⁹⁷ have been applied to demonstrate that dimensionality and matrix cues synergistically support maintenance of GSC stemness. More complex matrices can be fabricated by combining decellularized porcine or patient-derived brain matrix with low amounts of collagen, which better mimics the compositional complexity of the brain.^198,199 However, these matrices are limited by sample size and require processing steps that destroy the native protein structure.

Cells can be embedded into 3D hydrogels as tumourspheres or as homogenously dispersed single cells (FIG. 2b). Spheroids recapitulate the soluble cue gradients present in tumours, and spheroids with large diameters (>500 μm) exhibit a hypoxic and sometimes necrotic core.²⁰⁰ GSCs cultured as tumourspheres in serum-free medium better maintain stemness and heterogeneity than GSCs cultured as single cells in serum-containing medium, and they can be directly encapsulated into matrices.²⁰¹ Adherent cells can be grown as tumourspheres using a hanging drop culture²⁰² or microwells²⁰³ to aggregate cells into spheroids. Homogenous dispersion of single cells, which are typically encapsulated during matrix gelation, enables evaluation of single-cell morphology, proliferation and colony growth.^193,204 In matrices with large pores, liquid cell suspensions can be dropped onto dehydrated, hydrophilic scaffolds; the cells are then drawn into the bulk 3D matrix after rapid absorption. This approach allows the incorporation of cells into matrices with harsh fabrication chemistries, such as electrospun polystyrene or porous PCL.^194,195 Stromal cells can also be integrated into 3D matrices together with tumour cells, but with limited possibilities to control their spatial organization. Stromal cells strongly influence GBM cell behaviour; for example, GBM cells cultured with astrocytes and endothelial precursors in 3D HA–collagen matrices exhibit increased migration speed and resistance to STAT3 inhibition, as compared to GBM cell culture alone.²⁰⁵

HA-containing matrices can be fabricated by directly crosslinking the HA backbone^136,183, by complexing HA with polycations such as chitosan²⁰⁶, or by mixing or conjugating HA into hydrogel networks with collagen²⁰⁷, gelatin^45,208 or PEG.²⁰⁹ The nanoporosity (~100–200 nm mesh size) of crosslinked HA gels impedes cell squeezing, necessitates more cell-mediated matrix degradation and leads to slower invasion than matrices with large pores, such as collagen.^183,184,190 HA can also be mechanically incorporated into gelatin matrices with variable elastic moduli and growth factor concentrations. The specific combinations of modulus and growth factor differentially affect proliferation and invasion.²¹⁰ Using high-molecular weight HA, as compared to low-molecular weight HA in gelatin matrices leads to an increase in HA production by GBM cells and a decrease in cellular invasion, without changes in HA synthase or hyaluronidase protein expression.⁴⁵ The presence of HA in 3D models further induces resistance to the EGFR inhibitor erlotinib, mediated by CD44²¹¹, as well as altered RHAMM, HAS1, and HAS2 gene expression.¹²⁴ The effect of HA on resistance to erlotinib depends on the mutant status of EGFR, which can vary between patient-derived lines.²¹² Thus, the incorporation of HA into engineered TME models has revealed key mechanisms by which HA drives GBM progression.

Engineering gradients

Mechanical and biochemical ECM cues in the brain are often spatially organized, for example as gradients or localized hotspots. Spatial organization can be recreated by 2D substrate patterning using photolithographic and microfabrication techniques in combination with aqueous photochemistries.^213,214 For example, polydimethylsiloxane (PDMS) substrates can be patterned with different stiffnesses by generating stiff posts of defined shapes and sizes, which can be attached to the underside of a thin PDMS membrane. Fibroblasts and myoblasts cultured directly opposite the pillars on the flat upper side of the membrane experience the highest stiffness, and show a haptotactic response by migrating towards or along stiff features.²¹⁵ Patterning substrates with ECM or mechanical gradients can be used for high-throughput parameter space testing or to examine cell responses to brain-like haptotactic cues. For example, orthogonal patterning of a fibronectin and elastic modulus gradient on an HA hydrogel revealed that GBM cells spread and express oncogenic miRNA in a non-linear manner across the range of the gel.¹⁸² Patterning of 3D substrates is limited by the available patterning method. For example, microfluidic mixing of HA and gelatin precursor solutions with different concentrations results in 3D gelatin–HA gels with gradients of crosslinking density, in HA content and subsequently in cell density.²¹⁶ Interestingly, cells in these gels showed a biphasic MMP9 expression profile with increasing HA concentration. 3D gels can also be attached to a glass surface resulting in a non-linear stiffness gradient along the z-axis. Cells encapsulated less than 25 μm from the glass surface spread more and migrate faster than cells located > 500 μm from the glass surface independent of matrix density, demonstrating that distance from the glass substrate to the cells within the gel could be used to investigate mechanical effects on GBM.²¹⁷ Soluble cue gradients, including oxygen gradients and hypoxia, arise naturally in bulk 3D gels submerged in medium as a function of gel thickness. Cells seeded in 2 mm thick gelatin hydrogels are exposed to lower rates of nutrient transport and show a pro-angiogenic phenotype with increased VEGF and HIF-1 expression, as compared to cells cultured in 1 mm thick gelatin hydrogels.²¹⁸ Therefore, these TME models can be applied to elucidate the mechanisms by which spatial variation in mechanics, ECM composition and soluble cues influence tumour progression.

Engineering interfaces and topography

Semi-3D materials, often referred to as 2.5D materials, are characterized by a 3D topology arising from multiple 2D topologies. 2.5D systems combine the practicality of fabricating 2D features or patterns with the possibility to incorporate 3D-like constraints. In certain cases, these systems more faithfully recapitulate tissue architecture than ‘true’ 3D matrices. For example, the interface between the vascular basement membrane and the intraparenchymal ECM has been modelled by consecutively layering materials that are representative of the two regions (FIG. 2c). The bottom layer fabricated from Matrigel is analogous to the vascular membrane and the top layer of viscous, soluble HA is analogous to the parnechyma.²¹⁹ GBM spheroids seeded at the interface of the two layers show rapid, collective cell migration along the interface when the top layer includes highly viscous HA or viscous methylcellulose as compared to little invasion when the top layer does not include viscous HA or methylcellulose. Thus, the presence of an interface between a matrix layer and highly viscous solution is sufficient to guide cell invasion along vascular membranes. The migration speed of cells seeded between fibronectin-coated PA and crosslinked HA or crosslinked HA conjugated with the integrin-binding peptide RGD depends on the degree of ligand–receptor interactions between the cells and the interface, with more interactions slowing invasive migration speed.²²⁰ Semi-3D substrates resembling the brain intraparenchymal region can also be fabricated by layering ECM-producing astrocytes onto plastic to form a parenchyma-like substrate.²¹⁹ GBM invasion speed on astrocyte layers inversely correlates with the culture time of astrocytes, which may be a result of ECM accumulation or changes in astrocyte phenotype.

GBM cells rapidly invade along anatomical tracks, specifically in the perivascular space or on myelinated axons.¹⁸ Engineering models of anatomical tracks typically include a linear, topographical feature fabricated on a 2D surface or encapsulated in a 3D matrix. Confinement imposed by microchannels can recapitulate the linear migration and squeezing that cells exhibit when invading tight spaces along anatomical tracks. PA microchannels can be employed to independently modulate pore size and modulus, and have been used in our laboratory to show that matrix modulus and confinement synergize to promote rapid invasion.¹⁷⁹ Alternatively, nanofibres can be applied to study the effects of aligned topographical cues resembling the orientation of white matter tracts. Interestingly, aligned fibres strongly promote rapid, linear migration (FIG. 2d).^221–226 To decouple the surface chemistry from the fibre mechanics, electrospun fibres with a ‘core’ material surrounded by a ‘shell’ of a different material were fabricated. The core material determined the modulus, while the shell material determined the surface chemistry. Varying material combinations for the shell and core were employed to demonstrate that GBM cell migration and morphology are sensitive to both nanofibre modulus and ECM coating.²²⁵ The basement membrane composition and topographical features can be recreated within a 3D matrix by coating microfibres with Matrigel and embedding them in 3D matrices. Invading cells that encounter microfibers switch to an invasive mode and rapidly migrate along the fibres.²⁰⁷ ECM-coated nanofibres also modulate GSC stemness, with laminin-isoform-specific effects.¹⁹⁴ Thus, topographical cues strongly drive invasion, proliferation and resistance, which can be enhanced by other TME signals, such as ECM composition and increasing stiffness.

Interstitial fluid in engineered models

Little is known about how interstitial fluid flow and pressure direct GBM invasion. Interstitial fluid flow can be modelled by seeding hydrogel-encapsulated cells in a Boyden chamber. The top chamber is then filled with excess medium, which creates pressure-driven fluid flow through the membrane pores in parallel to cell migration (FIG. 2e). Using such a model, it could be demonstrated that the interstitial fluid flow activates CXCR4-dependent polarized cell migration in multiple GBM cell lines, including GSCs.^95,96 This CXCR-4 dependent invasion was confirmed in a mouse model, in which convection enhanced therapy was applied to control interstitial flow,⁹⁷ highlighting the clinical importance of fluid flow for tumour progression and convection enhanced therapy.^97,170

Microfluidic models with multiple cues

Adding more complexity to TME models improves physiological relevance, but typically increases the required labour and sacrifices throughput.²⁹ Microfluidic models can be made complex enough to facilitate construction of TME models with fluid flow, 3D ECM, spatial organization and stromal cell co-culture in a single platform, while allowing imaging, control of parameters, and high-throughput screening²²⁷ as well as achieving cost-effectiveness, compared to in vivo models. For example, a device with three parallel, adjacent channels has been developed to test the hypothesis that pseudopalisades form as migrating cells accumulate after a vaso-occlusive event (FIG. 2f).²²⁸ The outside channels contain flowing medium and the centre channel contains a 3D matrix with homogeneously encapsulated cells. Vaso-occlusion can be mimicked by stopping the flow through one channel, which results in a hypoxic gradient. GBM tumour cells migrate away from the occluded channel and form pseudopalisades, supporting the mechanistic hypothesis.

The versatility of microfluidic devices also allows the reconstruction of TME niches. In particular, perivascular niche models can be constructed using parallel, interconnected channels to spatially organize niche layers. GSCs incorporated into such a microfluidic perivascular niche model featuring endothelial cells and the spatial organization of a GBM tumour exhibit morphologies, stemness markers and CXCR4-dependent invasion similar to those observed in vivo (FIG 2g).²²⁹ Similarly, in a three-channel device with a tumour reservoir separated by a collagen matrix from an endothelialized vascular-like reservoir, GSCs are known to precede their differentiated counterparts in invasion. Moreover, GBM pro-invasive genes, including integrins α2 and β3, are upregulated in the presence of endothelial cells.²³⁰ Vascular homing can be studied using a microfluidic device, in which GSCs are encapsulated in a 3D microvascular network.²³¹ GSCs derived from the subtype of GBM tumours with high platelet-derived growth factor receptor A (PDGFRA) expression are particularly prone to vascular homing.

Microfluidic devices have also been developed for preclinical screening. Numerous wells can be included in a single device, seeded with tumourspheres and exposed to orthogonal gradients of chemotherapeutics and nutrients. These devices can serve as platforms for the optimization of drug efficacy and to predict therapeutic resistance.^232–234 However, how these results would translate to decisions for patient care remains unclear given the difficulty in validating in vitro results with patient outcomes. The efficacy and toxicity of chemotherapeutics are significantly influenced by multiple organ system functions, particularly by the liver metabolism. Intestine and liver models can be added to a GBM model in a microfluidic device to allow chemotherapeutic screening, while considering pro-drug absorption by an intestine-like lumen as well as metabolism by liver cells.²³⁵

Bioprinting

Bioprinting, or 3D printing of biomatrices and/or cells, can be applied to organize and fabricate 3D matrices and microfluidic models.^236,237 For example, patient-specific GBM models can be bioprinted using concentric rings of endothelial and patient-derived tumour cells encapsulated in a porcine brain-derived matrix (FIG. 2h).¹⁹⁹ Key tumour features, such as the hypoxia-induced necrotic core surrounded by psuedopalisades, were observed within the model. Importantly, printed tumours recapitulate clinically-observed patterns of tumour resistance to standard therapeutic treatments. The printing of patient-specific tumour models is limited by the sample size of the resected tumour; however, these results demonstrate the feasibility of incorporating a brain-derived matrix into printable bioinks in combination with patient-derived cell lines to test therapeutic responses. Similarly, bioprinted ‘mini-brains’ comprising a tumour-like, cell-dense region surrounded by a brain-shaped, macrophage-laden gel mimic the spatial organization of TAMs. The GBM cells in this model recruit macrophages and influence macrophage polarization; in turn, macrophages induce GBM invasion.²³⁸

Organoid models of growth and invasion

Instead of recapitulating the complex brain matrix by controlled fabrication, cells can also be seeded into a matrix and stimulated to spontaneously develop into an organoid. To generate GSC organoids, patient tumour samples can be seeded directly into Matrigel suspended in medium. The suspended tumour cells grow into ‘tumours’ with diameters of 5–10 mm over 5–6 months.²³⁹ In contrast cell isolation methods in which the matrix is degraded and cells are disassociated, this method better preserves patient cell–matrix interactions and tumour heterogeneity, including the proportion of GSCs relative to differentiated cells found in the original patient tumour. During organoid growth, a GSC-rich hypoxic niche is formed at the centre of the organoid, which is surrounded by more rapidly dividing cells. Compared to cells cultured in spheroids, cells in organoids better mimic patient tumour phenotype and heterogeneity in orthotopic xenograft models as well as therapeutic resistance in vitro. Similarly, cerebral organoids with organized, differentiated brain features have been developed for other disease models.²⁴⁰ These approaches could also be combined to study GBMs. For example, in GSCs, which were seeded in engineered human nervous tissue generated from pluripotent stem cells, the expression of more than 100 genes was upregulated by interactions of GBM cells with stromal cells, many of which relate to ECM remodelling²⁴¹. Therefore, organoid models and engineered tissue can be applied to capture the complexity of tumour TMEs; however, their fabrication is time-intensive and they are difficult to reproduce. The benefits of complexity often do not outweigh the costs.

Opportunities for engineered models

Engineered GBM TME models have already provided a wealth of information about the function of the TME in GBM progression, including context-dependent mechanisms of GBM invasion and therapeutic resistance. With improved accuracy and (patho)physiological relevance, GBM TME models will play an important role in the preclinical and clinical pipeline (FIG 3); for example, platforms incorporating patient-specific tumour samples may eventually aid in predicting therapeutic response and for the tailoring of treatments.^{228,232,239,242} Drug responses are currently just as or more robustly predicted by molecular subtype, DNA methylation status and patient age than by in vitro testing. Furthermore, the limited treatment options in GBM arguably do not yet necessitate complex optimization strategies.^1,243,244 However, validated and reliable engineered models could greatly improve preclinical drug testing. Established mouse models for in vivo screening have already been incorporated as secondary endpoints in GBM clinical trials²⁴⁵; however, the time required for model development hinders timely translation into personalized therapies. Engineered TME models would allow therapy screening at shorter timescales. Furthermore, the development of microfluidic models of drug permeability across the BBB could be very valuable for evaluating drug delivery to the central nervous system.²⁴⁶ Such models are already being developed, but require additional validation and standardization.²⁴⁷

Figure 3. Glioblastoma microenvironment models in the preclinical and clinical pipeline.

Red dashed lines indicate stages at which engineered models are or could be used. a) Engineered tumour microenvironments (TMEs) have been widely employed as research platforms to investigate the TME and they can be used to identify therapeutic targets. b) With refinement, these platforms can serve as a basis for precision medicine using patient-specific cells and/or matrices. c) Images of tumours from patients can be used to generate mechanically-matched patient-specific models of the tumour and brain anatomy for surgical planning and training. d) After surgical resection, engineered TMEs can aid in maintaining heterogeneity during culture for patient-specific treatment validation. The cells can be selected by molecular profiling and histological analysis.

The translation of patient-specific anatomy to engineered models is also becoming achievable owing to advances in 3D printing technologies.²³⁶ Full-scale brain models can be generated from patient MRI scans and have proven useful in pre-surgical planning, teaching and training.²⁴⁸ For example, gelatin-based brain models have realistic mechanical properties and can be used for practicing gross resection without damaging intact tissue.²⁴⁹ Further inclusion of a 3D printed skull enables surgeons to practice skull cutting and how to access the tumour site without damaging tissue.²⁵⁰ Printing of patient-specific anatomical features combined with patient-derived cells and matrix may better recapitulate the gross tumour, facilitating at-scale studies of the TME. Such models could be useful for studying the influence of interstitial fluid on therapeutic delivery, for example, on drug release from Gliadel wafers or convection enhanced delivery.^114,170

Machine learning strategies can also be applied to GBM research. For example, algorithms can be used to extract functionally predictive information about the TME from MRI images. In particular, machine learning-based parameterization of contrast enhancement in MRI images correlates with gene expression of distinct biological processes, such hypoxia, starvation, matrix remodelling and endothelial permeability.²⁵¹ Furthermore, image features can be correlated with tumour subtype and patient survival.^252,253 Patient-specific MRI data can then be combined with other patient characteristics, such as age and Karnofsky Performance Score, to improve diagnosis before surgical resection is performed.²⁵⁴ Machine learning has also been explored to improve tumour segmentation.²⁵⁵ This is particularly important for surgical planning, but could also be applied for early diagnosis and therapy selection. The information derived from machine learning algorithms could be combined with other TME modelling technologies to improve their accuracy.

Perspective and conclusions

The TME has demonstrated potential as a therapeutic target for GBM treatment owing to its impact on tumour progression. Engineered microenvironments allow the investigation of cell responses in the context of the TME and thus facilitate rapid hypothesis testing and screening. However, challenges remain. In particular, the minimal model components necessary to accurately recapitulate in vivo mechanisms need to be determined, and the accuracy of models needs to be validated. It remains unclear which of the numerous ECM formulations used in engineered models meet these minimal requirements. A reductionist approach in developing TME models is useful to mimic in vivo GBM cell behaviour while avoiding unnecessary costs and complexity. Validation ensures that in vitro discoveries generate useful predictions of clinical relevance. Validation strategies have not yet been fully standardized, but generally fall into two categories. First, it has to be demonstrated that the physical parameters of the model, such as composition and mechanics, closely match the ones of brain, to make the model predictive of in vivo behaviour. Second, as a measure of model accuracy, cell phenotypes, such as migration, morphology, relative gene expression and chemosensitivity should be similar to the in vivo phenotype. Ideally, it should further be verified that tumour progression in engineered models is driven by similar biochemical mechanisms as in vivo (for example, signalling pathways governing drug resistance), although this is currently rarely done. An iterative design cycle could be created, in which TME models are systematically tested, and the mechanistic and phenotypic predictions are checked against the in vivo response to refine the model and to improve its predictive power.

Practical challenges that limit customizability and complexity include limited throughput and the need for composite fabrication techniques. Co-culture of GBM cells and stromal cells poses particular challenges, such as medium incompatibility, unmatched proliferation rates and long-term viability of primary stromal cells. Similarly, the inclusion of patient-derived cells or matrix in engineered models faces several challenges. Tumour matrix is difficult to obtain in large quantities, and the acclimation of tumour cells to cell culture can alter their phenotype. However, these challenges can certainly be addressed in the future, and engineered models offer the opportunity to rapidly and precisely dissect mechanisms of GBM progression, to accelerate clinical testing and to provide a platform for precision medicine.