Exogenous matrix‐free biomanufacturing of glioblastoma organoids enables autonomous assembly of neurovascular‐like structures and extracellular matrix

Abstract

Current glioblastoma (GBM) models often rely on chemically undefined exogenous extracellular matrices (ECMs) and serums that limit understanding of autonomous cellular behaviors driving tumor progression. Here, we present an exogenous ECM‐free biomanufacturing process for GBM organoids (GBOs) that recapitulate pathological development relative to size and spatial distribution. These GBOs exhibit a conserved glioma signature established from The Cancer Genome Atlas clinical datasets. Trajectory analyses of neurovascular unit (NVU) zonation, cell‐type differentiation, and basement membrane converged with clinical benchmarks highlighting a critical developmental milestone at the 2 mm diameter stage. Immunofluorescence revealed a matrix‐priming cascade of tenascin‐C (TNC) at 1 mm diameter and fibronectin (FN) at 1.5 mm diameter to provide the necessary biochemical cues and biophysical assembly for endothelial cells at 2 mm diameter. Lastly, the 2 mm diameter GBOs displayed a distinct vascular‐like area characterized by capillary‐scale lumen distributions (5–15 μm) and peak junctional complexity, similar to in vivo capillary beds. However, transcriptional divergences in arterial and proteoglycan markers suggest biochemical cues and morphology are insufficient for full physiological maturation without confirmed internal hemodynamic shear stress. Despite these results, our GBOs serve as a high‐fidelity structural framework for modeling GBM NVU‐like regions that may be used as a translatable platform for TNC‐ and FN‐targeted therapies designed to enhance drug delivery across the NVU.

Glioblastoma stem cells that are capable of self‐renewing and differentiating were bioengineered to become organoids with endogenous tumor microenvironment. The organoids exhibited an autonomous assembly of the necrotic core, infiltrating rim, and neurovascular‐like features guided by extracellular matrix deposition.

Translational Impact Statement.

This work presents a novel bioengineering approach to overcome limitations in modeling glioblastoma (GBM) interactions with neurovascular unit and extracellular matrix components by biomanufacturing GBM organoids (GBOs). By aligning GBO development with clinical progression baselines from TCGA transcriptomic data, we established a platform that recapitulates critical GBM developmental milestones. This system reduces reliance on undefined animal‐derived components and offers a reproducible, high‐throughput model for identifying therapeutic targets—specifically, tenascin‐C and fibronectin—ultimately providing a human‐centric proxy for preclinical drug validation.

1. INTRODUCTION

Gliomas are the most common malignant primary brain tumors in adults, arising from mutations in glial cells. ¹ Glial cells are non‐neuronal cells, including astrocytes and oligodendrocytes, that support and insulate the central nervous system (CNS). ^{2 , 3 , 4 , 5} Recent multi‐omic advancements prompted the World Health Organization to restructure CNS tumor classifications in 2021, shifting toward molecularly defined diagnostics. ^{6 , 7} Gliomas are categorized into four grades based on malignancy, ranging from slow‐growing Grade 1 tumors to highly aggressive Grade 4 glioblastoma (GBM). ¹ GBM is primarily characterized by an isocitrate dehydrogenase (IDH)‐wildtype status and arises de novo, whereas IDH‐mutant high‐grade gliomas typically progress from lower‐grade gliomas (LGGs). ^{5 , 6 , 7 , 8} Despite aggressive treatment, the median prognosis of GBM remains at 12–15 months with a 2–3 year rate of survival. ^{1 , 9} These poor outcomes underscore a critical need for physiologically relevant models that capture the complex GBM tumor microenvironment (TME).

The GBM TME is defined by a heterogeneous landscape of vasculature, extracellular matrix (ECM), a necrotic core, and an infiltrating rim. ^{10 , 11 , 12 , 13} Given that astrocytes are the primary glial cell of origin for many gliomas, the neurovascular unit (NVU) is an essential component of any high‐fidelity mode. ^{14 , 15 , 16} The NVU is a highly organized network designed to regulate nutrient delivery and maintain the blood–brain‐barrier (BBB). ^{17 , 18 , 19} This system spans a structural hierarchy: arteries and arterioles deliver oxygenated blood, veins and venules facilitate deoxygenated export, and a dense capillary bed allows for nutrient exchange. ^{20 , 21 , 22 , 23 , 24 , 25} Structurally, these vessels transition from thick layers of smooth muscle cells (SMCs) in larger vessels to a single layer of endothelial cells (ECs) enveloped by pericyte cells (PCs), and astrocyte cell (AC) end‐feet in capillaries. ^{26 , 27} This EC–PC–AC interface, often referred to as the BBB, is stabilized by tight junction proteins and a self‐fabricated basement membrane rich in proteoglycans and fibrous ECM. ^{4 , 20 , 28 , 29 , 30 , 31 , 32 , 33} In pathology, vascular leakage and ECM remodeling trigger cancer‐associated fibroblast (CAF) activation and “leaky” angiogenesis—hallmarks that are difficult to replicate in vitro. ²⁵

Organoids have emerged as 3D tissue‐like modeling systems that bridge the gap between simplistic 2D mono‐cell cultures and inaccessible human tissue. ^{34 , 35} Glioblastoma stem cell (GSC)‐derived organoids (GBOs) are typically generated by encapsulating GSCs in exogenous ECM hydrogels and differentiating them via serum supplementation. ^{36 , 37 , 38 , 39 , 40 , 41} However, the current reliance on these media serums and exogenous matrices masks the cells' innate ability to fabricate their own microenvironment. ^{42 , 43 , 44} These undefined conditions limit reproducibility and crucially hinder our understanding of how essential structures like the NVU develop during tumor progression. ^{16 , 17 , 45} Furthermore, the lack of standardized biophysical cues, such as shear stress and geometric control, often results in models that fail to recapitulate the spatial organization seen in patient pathology.

In this study, we hypothesized that by mimicking physiological shear stress and geometric constraints, GSC‐derived organoids would self‐assemble a structural and transcriptomic environment that recapitulates NVU‐like regions and their associated ECM, mirroring the developmental trajectories observed in patient glioma progression. To test this, we developed a GBO model that resembles patient pathology by focusing on the co‐development of the endothelial networks and their associated ECM. We first established a transcriptomic baseline of human glioma progression (non‐tumor or “Normal” to LGG to GBM) using The Cancer Genome Atlas (TCGA) data to identify a “conserved glioma signature” for the NVU‐like regions and their associated ECM. Using a patient‐derived xenograft (PDX) line, we employed GSC‐enriching media under controlled geometric and shear conditions to grow GBOs up to 4 mm in diameter. We demonstrate that these GBOs not only replicate the patient transcriptomic signature but also mimic the structural zonation, spatial distribution, and morphometric scales similar to the human NVU. Ultimately, by providing a high‐fidelity, patient‐derived platform that captures the structural and transcriptomic complexities of the NVU, this work establishes a robust framework for investigating tumor–matrix interactions and provides a scalable tool for the preclinical screening of therapeutics targeting the GBM vascular microenvironment.

2. RESULTS

2.1. Transcriptomic profiling of glioma progression establishes a pathological baseline for NVU‐like regions and associated ECM remodeling

To establish a high‐fidelity baseline for glioblastoma organoid (GBO) maturation, we first performed transcriptomic profiling of clinical glioma progression. Publicly available patient data from The Cancer Genome Atlas (TCGA) was acquired and curated to a high‐confidence subset of 35 cases to ensure balanced downstream comparisons between clinical and organoid datasets. ^{46 , 47 , 48} Principal component analysis (PCA) revealed a distinct developmental axis, with LGG cases exhibiting a spectrum of transcriptomic identities between healthy tissue and malignancy. To capture this transition, LGG samples were subcategorized into three groups based on their molecular proximity to clinical benchmarks: LGG‐Normal (LGG‐N), LGG‐GBM (LGG‐G), and a transitional LGG group (Figure 1a).

FIGURE 1.

Transcriptomic benchmarking of clinical glioma progression. (a) Principal component analysis (PCA) of The Cancer Genome Atlas clinical data, illustrating the subcategorization of LGG‐normal (LGG‐N) and LGG–GBM (LGG‐G) cohorts. Venn diagram (b) and volcano plot (c) highlighting the conserved differentially expressed genes (DEGs) across glioma grades. Heatmap (d) and enriched gene ontology (EGO) analysis (e) showing upregulation in tissue genesis and embryonic skeletal morphogenesis, with concomitant downregulation in GABA‐ergic neuronal pathways. (f) CNET network plot identifying central hubs linking vasculogenesis and collagen metabolic processes. GBM, glioblastoma; LGG, low grade glioma.

Differential expression analysis established a conserved glioma signature characterized by a clear molecular hierarchy (Figure 1b,c). This pathological profile was defined by a significant upregulation in gene ontologies (GOs) associated with embryonic skeletal development and oxidative phosphorylation, alongside a simultaneous suppression of mature neuronal pathways, particularly GABA‐ergic signaling (Figure 1d,e). While global tissue genesis pathways dominated the top‐tier enrichment scores, network analysis confirmed our central hypothesis: glioma progression is supported by a coordinated remodeling of NVU‐like structures and ECM. Specifically, we identified VEGFA and FN1 as critical network hubs connecting vasculogenesis and collagen‐metabolic processes based on our previous characterizations of the organoid ECM (Figure 1f). ^{16 , 45} These defined patient signatures served as the baseline for evaluating the structural and cellular fidelity of our biomanufactured GBO models.

2.2. GBOs recapitulate capillary‐specific zonation trajectories observed in patient disease progression

Evaluation of vascular‐like fidelity of GBOs to patient disease progression led to comparison of transcriptomic trajectories of patient disease progression against organoid development across a unified five‐stage scale. While GBOs exhibited a divergence in arteriole and large vein markers, they demonstrated concordant trends in capillary, venule, and large artery trajectories (Figure 2a). The maturation of these structural zones is supported by the coordinated differentiation of the constituent NVU cells: ECs, ACs, PCs, SMCs, and fibroblasts (Figure 2b). Analysis of cell‐type‐specific trajectories revealed a high degree of overlap between GBO and patient signatures for ACs, fibroblasts, and PCs/SMCs, with peak alignment occurring at the 2 mm GBO developmental stage. Similarly, markers associated with cell‐to‐cell adhesion and fibrous ECM in the basement membrane showed significant overlap at this 2 mm milestone (Figure 2c). The convergence of the cell type, adhesion, and fibrous ECM suggests a synchronized structural maturation at the 2 mm GBO stage.

FIGURE 2.

Recapitulation of vascular zonation, cell‐type, and basement membrane trajectories. (ai) Longitudinal transcriptomic trajectories of vascular zones (ii) large arteries/arteries and arterioles, (iii) capillaries, and (iv) venules and veins/large veins in patient data (blue) versus GBO development (red). (b) Cell‐type specific trajectories for astrocytes (AC), endothelial cells (EC), fibroblasts, and pericytes/smooth muscle cells (PC/SMC), showing peak convergence at the 2 mm GBO stage. (c) Expression profiles for basement membrane components: Adhesion markers, fibrous extracellular matrix (ECM) and proteoglycans. Divergences in arterioles, large veins/veins, endothelial cells, and proteoglycans are likely due to a lack of internal biophysical cues. Data represented as mean relative z‐score ± SEM. Created in BioRender. Avera, A. (2026) https://BioRender.com/7xw5vxp.

The observed divergences in EC identity, proteoglycan deposition, and arterial zonation likely stem from the absence of controlled hemodynamic shear stress (HSS). ^{49 , 50 , 51 , 52} HSS occurs from the tangential forces applied by the blood flow and is vital in EC differentiation and maintenance for vascular‐like stability. ⁵³ While GBOs exhibited significant upregulation in fibrous ECM and cell–cell adhesion markers (e.g., JAMs, occludin, and claudin), proteoglycans associated with cell–ECM adhesion were notably downregulated compared to clinical benchmarks. These results suggest that while the GBO platform provides the necessary biochemical cues for microvascular maturation and NVU cell recruitment, additional biophysical cues within the GBO may be required to fully recapitulate the mature arterial and venous identities observed in human pathology.

2.3. Temporal maturation of GBOs promotes co‐localization of pathological ECM and endothelial cell network assembly

Given that peak molecular correspondence for fibrous ECM and NVU‐constituent cells occurred at the 2 mm stage, we investigated the spatial co‐localization of the NVU‐like regions and its associated matrix during GBO maturation. Previous characterization of these GBO models demonstrated global expression of elastin, collagen IV, and collagen VI in 2 mm GBOs, contrasting with the minimal expression observed in initial spheroids. ⁴⁵ As collagen IV is a primary basement membrane component and established marker for EC localization, we sought to determine which ECM components would be spatially expressed and co‐localized with ECs. Thus, we focused on the spatial relationship between CD31+ (PECAM1) ECs and the fibrous ECM components fibronectin (FN) and tenascin‐C (TNC), which modulate adhesive and anti‐adhesive cellular behaviors, respectively (Figure 3).

FIGURE 3.

Spatial distribution and temporal cascade of the neurovascular unit‐like extracellular matrix (NVU‐ECM) niche. (a–d) Representative average intensity projections of confocal multilayer images showing GBO maturation from 0.6 mm (spheroid) to 2 mm GBO. Samples are stained for DAPI (blue), tenascin‐C (TNC, green), fibronectin (FN, red), and CD31 (endothelial marker, yellow). (b) Emerging TNC expression surrounding the core at 1 mm. (c) Peripheral deposition of FN at 1.5 mm. (d) Organized co‐localization of CD31 and FN and nascent endothelial clusters in the vascular‐like area at the 2 mm milestone, indicating a matrix‐primed vascular assembly. Scale bars: 100 μm.

While 0.6 mm spheroids exhibited negligible TNC, FN, or CD31 expression (Figure 3a), GBO maturation triggered a distinct spatial and temporal cascade. TNC expression emerged surrounding the core region as early as the 1 mm stage (Figure 3b), followed by the peripheral deposition of FN and the emergence of nascent endothelial clusters at the 1.5 mm stage (Figure 3c). Crucially, significant co‐localization of FN and CD31 was not observed until the 2 mm developmental stage, where the signals overlapped within the NVU‐like regions of the organoid (Figure 3d). These findings suggest that the self‐fabrication of a fibronectin‐rich matrix by GBO‐resident cells precedes endothelial recruitment, potentially serving as a structural scaffold for subsequent vascular assembly.

2.4. GBO vascular‐like architectures exhibit morphometric distributions mirroring clinical neurovascular unit dimensions

While the presence of endothelial cells and ECM suggests biological potential of an NVU, physiological relevance is defined by the spatial organization of vessel hierarchies categorized by lumen diameter, spacing, and branching. In accordance with the spatial trends observed in protein expression, 2 mm GBOs exhibited a distinct tri‐zonal architecture: an infiltrating rim, a vascular‐like area (VLA), and a hypoxic/necrotic core (Figure 4a). The rim region (TNC⁺/FN⁻/CD31⁻) aligns with previously characterized hyaluronic acid (HA)‐rich regions identified via CD44 expression. ⁴⁵ The TNC⁺/HA⁺ rim suggests that the combination of anti‐adhesive TNC and HA facilitates high migratory and infiltrative potential at the organoid periphery. Given the ~200 μm limit of oxygen diffusion, the observed rim thickness of 153–180 μm suggests a structurally defined metabolic boundary. ²¹ Immediately interior to the rim, we identified a substantial VLA (~750–1013 μm width) characterized by TNC^low/FN⁺/CD31⁺ profile, followed by a presumed hypoxic/necrotic core (TNC⁻/FN⁻/CD31⁻) with a radius of ~190–220 μm. ¹⁶ Analysis of CD31⁺ architectures revealed that the VLA contained the highest vessel branching density and lowest fractional lumen spacing (Figure 4b,c). These metrics indicate a highly organized, dense network assembly that reflects the “weblike” capillary beds found in vivo.

FIGURE 4.

Morphometric distribution and regional zonation of vascular‐like architectures. (a) Schematic and representative image of the tri‐zonal architecture in a 2 mm GBO: Rim, vascular‐like, and core. (b) Normalized lumen count, (c) normalized lumen spacing, and (d) normalized junction density across regions; the vascular region exhibits a local minimum in spacing and a peak in junctional complexity. (e) Histogram of lumen size distribution via minimum Feret diameter, highlighting a predominance of capillary scale (5–15 μm) architectures. Data represents n = 4 biological replicates; outliers removed via Grubbs' test; *p ≤ 0.05; statistical trends were non‐significant by one‐way repeated measures ANOVA (p > 0.05) in (d).

Lumen quantification via minimum Feret diameter revealed a broad distribution of vascular‐like scales within the GBO (Figure 4d). The majority of lumens (5–10 μm range) matched the expected dimensions of human capillaries, while larger structures fell within the arteriole and venule ranges. Notably, no lumens exceeded 105 μm in diameter. When integrated with the zonation trajectories, these morphometric distributions confirm that the GBO platform supports the biophysical assembly of multi‐scale vascular‐like architectures with high capillary and venule concordance.

2.5. Macromolecular diffusion of 3 kDa fluorescent tracer (FITC)‐dextran through the NVU‐like regions of the GBOs

While drug testing was outside the scope of this work, it was necessary to perform a preliminary validation of diffusion throughout the GBOs and within the self‐assembled capillary‐like structures. The functional transport of the low molecular weight (3 kDa) fluorescent tracer (FITC‐dextran) was chosen because of the low molecular weight (194.15 g/mol) of the standard chemotherapy for GBM, temozolomide. ⁵⁴ The diffusion coefficient was determined to be 0.15 μm²/s and the experimental transport gradient closely fit Fick's second law of transport modeled in spherical coordinates (Figures 5a, S2). While empty or necrotic regions in the GBOs are evident as indicated in the top right of the sample GBO, there was clear transport of FITC‐dextran within capillary‐like lumens ranging between 9 and 12 μm diameter (Figure 5b). The functional transport of FITC‐dextran within the GBOs is apparent, but the determined effective diffusion coefficient was two orders of magnitude lower than what has been previously described in in vivo mouse brains. ⁵⁵ The slow diffusion may be due to the densely packed outer rim that protects the expanding inside of the GBO. The precise introduction of physiologically relevant HSS within these functional capillary‐like structures may encourage more vascular branching through the protective rim region to facilitate more in vivo‐like transport properties.

FIGURE 5.

Functional transport properties of FITC‐dextran in 2 mm organoids. GBOs were stained with FITC‐dextran (3 kDa) for 0, 1, and 6 h, counterstained with Hoechst, whole‐mount cleared, and imaged by multiphoton microscopy. (a) Representative maximum intensity projection of a 6 h stained GBO volume. The lateral orthogonal side panels represent a 50 μm z‐stack acquired at 0.5 μm step resolution across the core boundary. Scale bar: 100 μm. (b) Spatiotemporal mass transport profiles as a function of radial distance from the organoid core (r = 0 μm) to the tissue periphery. Experimental micro‐environmental intensity gradients were fitted to a center‐out spherical form of Fick's Second Law, yielding an effective diffusion coefficient ${\mathcal{D}}_{\mathrm{eff}}$ = 0.152 μm²/s. n = 3 biological replicates for each condition; shaded bands refer to the 95% confidence intervals. (c) High‐resolution regional zonation analysis within the neurovascular‐like region (NLR) at 6 h showing: (i) raw FITC‐dextran channel, (ii) intensity‐weighted skeletonized centerlines overlaid on a dim binary luminal mask, and (iii) a local vessel diameter heatmap. Scale bar: 10 μm.

3. DISCUSSION

The advancement of translational medicine is fundamentally dependent on patient‐derived modeling systems that can faithfully replicate the unpredictable microenvironment of diseased tissues such as glioblastoma (GBM). Early iterations of GBM organoids (GBOs) were predominantly generated using undefined serum and exogenous extracellular matrix (ECM) hydrogels to support differentiation and structural organization. ^{36 , 39 , 40 , 56 , 57 , 58 , 59} While these methods provided initial 3D frameworks, they often masked the inherent capacity of glioma stem cells (GSCs) to spontaneously synthesize their own tumor tissue—a process that appears to recapitulate de novo neurogenesis and vasculogenesis from a primitive state. ^{60 , 61 , 62} Furthermore, the reliance on undefined products introduces high batch‐to‐batch variability, creating significant vulnerabilities in biomanufacturing reproducibility and translational reliability. ^{35 , 36 , 38 , 63 , 64 , 65 , 66} To address these limitations, we redesigned the GBO biomanufacturing process to its simplest, most defined form, allowing for a characterization of development that parallels human pathology relative to size and spatial distribution. Our previous results identify the 2 mm developmental stage as a critical maturation milestone, exhibiting the peak upregulation of ECM transcriptomics and autonomous matrix fabrication. ⁴⁵ This maturation profile aligns with the conserved patient glioma signatures identified in this study, supporting a developmental cascade where self‐fabricated matrix precedes and guides structural assembly. Ultimately, by aligning structural zonation and lumen scale distribution with clinical benchmarks and oxygen diffusion limits, this platform provides a high‐fidelity framework that addresses the limitations of chemically undefined systems.

Establishing a conserved glioma signature and a developmental cascade within patient data was critical for ensuring the bench‐to‐bedside translatability of the GBO model. It is important to consider that prior to the 2021 World Health Organization (WHO) reclassification of CNS tumors, many gliomas were diagnosed primarily through tissue morphology and a limited panel of molecular markers. ^{6 , 7 , 67} Consequently, The Cancer Genome Atlas (TCGA) datasets from 2019 likely contain numerous low‐grade glioma (LGG) diagnoses that, by current molecular standards, would be reclassified as GBM. ⁶⁷ By subcategorizing the LGG cohort into groups such as LGG‐GBM based on transcriptomic proximity, we effectively modernized the clinical baseline to reflect current diagnostic realities. Despite these historical shifts in nomenclature, a core glioma signature—defined by the suppression of mature neuronal pathways and the activation of embryonic‐like growth and oxidative phosphorylation—remained conserved across all clinical grades and our GBO models. This conservation provides a robust molecular bridge, allowing for meaningful translation between retrospective clinical data and modern bioengineered models.

Leveraging this molecular bridge, we characterized the temporal maturation of our GBOs to determine if the conserved patient signature manifests as a coordinated developmental cascade of the neurovascular niche. Our results identify a critical developmental milestone at the 2 mm stage, where transcriptomic trajectories for capillary zonation, NVU cell differentiation, and fibrous ECM deposition converge with clinical benchmarks. This timeline suggests an ECM priming mechanism where early emergence of TNC at 1 mm establishes an anti‐adhesive migratory environment that precedes the structural stabilization offered by FN at 1.5 mm and beyond. The scale‐dependent transition is directly supported by transcriptomic profiling of core GSC markers and ECM markers TNC and FN1 (Figure S1). Cells exhibit high GSC uniformity (OLIG2, POU5F1, and PROM1 expression) expression under standard culture conditions and as spheroids, while 1 mm GBOs show activation of TNC followed by high upregulation of FN1 in 2 mm GBOs. While this sequential deposition of TNC and FN by stem cells is well studied in GBM and other diseased tissue (e.g., heart, lung, and breast), it has rarely been recapitulated in vitro. ^{68 , 69 , 70} The GBO system presented here successfully models the active GSC response to the emergence of a hypoxic/necrotic core. This response drives a remodeling and priming of local ECM through sequential TNC and FN deposition, ultimately facilitating the endothelial assembly observed at the 2 mm stage. By correlating these molecular events with specific morphometric distributions, our GBOs offer a unique glimpse into the self‐organizing kinetics of the GBM microenvironment without the confounding influence of exogenous scaffolds.

The evidence presented here focuses on the biophysical and transcriptomic organization of the GBO. To transition from this structural baseline to a functional platform, future studies will need to address the dynamic physiological components of the NVU. Moreover, discrepancy exists between transcriptomic profiles and protein‐level organization, particularly regarding PECAM1 (CD31) counts in our data suggest a critical detection threshold limit in the bulk profiling of heterogeneous organoids. Low abundance cellular lineages (i.e., emerging endothelial populations) may remain transcriptomically “silent” in bulk averages despite forming robust, spatially defined protein structures. Tracer spread within capillary‐like lumens followed Fick's second law, giving an effective diffusion coefficient (${\mathcal{D}}_{\mathrm{eff}}$) of 0.15 μm²/s for FITC‐dextran (3 kDa). A direct quantitative comparison of ${\mathcal{D}}_{\mathrm{eff}}$ to that in human tissues is not possible because such values, to the best of our knowledge, have not been reported. We therefore contextualize our result against the best available surrogate benchmarks. The closest macromolecular reference is Thorne and Nicholson's in vivo work in rat neocortex, where 3 kDa dextran in the extracellular space resulted in ${\mathcal{D}}_{\mathrm{eff}}$ = 54 μm²/s, which is more than two orders of magnitude higher than what we observed in our GBOs. ⁵⁵ We note, however, that they injected the dextran into rat parenchyma and measured diffusion in the extracellular space, whereas our fit comes from radial spread in the organoid vascular network. On the human side, Vargová et al. measured extracellular space volume fraction and tortuosity in resected glioma slices using tetramethylammonium, a much smaller probe. ⁷¹ In GBM, they found a larger extracellular volume (α ≈ 0.47 vs. ~0.24 in control cortex) and higher tortuosity (λ ≈ 1.67 vs. ~1.55), so high‐grade human glioma tissue already poses more diffusion resistance even for small ions. A 3 kDa dextran would likely face greater hindrance. Zámecník et al. linked increased tortuosity in high‐grade human astrocytic tumors to extracellular matrix deposition, particularly tenascin, implicating barriers beyond luminal geometry alone. ⁷² Together, these studies support additional hindrance in GBOs from the densely packed rim, incomplete endothelial junctions in nascent vascular channels, and absence of hemodynamic shear stress (HSS). Future iterations of GBOs must consider the integration of microfluidics or perfusion technology likely between the 1 and 2 mm developmental stage of the GBO to encourage stronger junctions between lining ECs for branching through the densely packed rim. ^{14 , 73 , 74 , 75 , 76} These controlled fluidic technologies will allow the GBOs to move beyond modeling the anatomy of the GBM microenvironment toward a truly functional, human‐centric platform for preclinical drug validation and personalized medicine.

The spatiotemporal coordination of TNC and FN observed in our GBO development system offers a compelling case for the development of ECM‐targeted therapies. Current GBM treatments often fail due to the highly invasive nature of GSCs at the tumor periphery and the inability to diffuse through the BBB. Our model suggests that GSCs prime the environment specifically with anti‐adhesive (TNC) to adhesive (FN) matrix switch to facilitate invasion and subsequent neovascularization. Consequently, utilizing these GBOs as a high‐throughput modeling system would accelerate the screening of small molecules or antibodies designed to disrupt TNC/FN scaffolding, potentially depriving the tumor of its ability to prime the neurovascular niche.

4. MATERIALS AND METHODS

4.1. Cell culture

The human patient derived xenograft (PDX) cell line JX6 was kindly provided by Dr. Yancey Gillespie (The University of Alabama at Birmingham). JX6 is a classical subtype of GBM with amplified variant III mutation (exons 2–7 deleted) of EGFR, CDKN2A homozygous deletion, and unmethylated MGMT promoter. JX6 was cultured in neural stem cell expansion medium (NBE) composed of Neurobasal™‐A medium (10888022, Gibco™) supplemented with 0.5× GlutaMax™ (35050061, Gibco™), 1× B‐27™ Supplement minus vitamin A (12587010, Gibco™), 1× penicillin–streptomycin (30‐002‐CI, Corning®), 2.5 μg/mL amphotericin B (30‐003‐CF, Corning®), 20 ng/mL recombinant human epidermal growth factor (100‐26, Irvine Scientific), 20 ng/mL recombinant human fibroblast growth factor‐basic 154/FGF‐2 (100‐146, Irvine Scientific), and 8 μg/mL heparin (AK3004‐1000, Akron). ^{77 , 78} Cells were expanded in standard tissue culture polystyrene flasks at a seeding density of 10⁵ live cells/mL and incubated at 37°C with 5% CO₂ cell culture conditions. Cells were fed as needed, every 2–3 days, until ready for passage. JX6 cells were dissociated with 1 mL of Accutase® (A6964, Sigma‐Aldrich) for 3–5 min at room temperature.

4.2. Glioblastoma organoids

GBOs were biomanufactured as previously published. ^{16 , 45} In short, the JX6 GBM tumor cells individually were expanded and seeded into a cell‐repellent surface U‐bottom shaped 96‐well plate (U96) (650979, Greiner) at a seeding density of 10⁴ live cells/100 μL media/well in the same NBE and culture conditions described in the cell culture methods. This initial seeding into a U96 was necessary to induce spheroid formation until the spheroids reached a critical threshold size (CTS) of 400–600 μm. After the GSC spheroids reached the CTS, they were transferred into a 100 mL dimpled bioreactor with an internal impeller (CL2‐1450‐100, Chemglass) at a sublethal shear stress of 0.5 Pa (120 rpm) induced by a magnetic stir plate (CLS‐4100‐A6, Chemglass) in the same NBE and cell culture conditions. No other cell culture media or ECM scaffolds were utilized throughout the entirety of GBO development.

4.3. Patient data from the cancer genome atlas

In short, patient data was retrieved from the Genomic Data Commons (GDC) by generating a cohort to extract data from The Cancer Genome Atlas (TCGA) projects labeled “TCGA‐LGG” and “TCGA‐GBM” projects. Open access data for STAR—Counts from the Illumina platform with specifications for solid tissue for both normal and primary tumors were downloaded then organized according to the metadata. This resulted in 438 total patient data files. In‐house R (v4.5.1) was utilized to perform differential gene expression analyses using the DESeq2 R package (v1.48.2) following the standard workflow as previously published. ^{79 , 80} Principal component analysis (PCA) was performed after performing variance stabilizing transformation on the DESeq data and visualized using ggplot2 R Package (v3.4.4). Initial PCA revealed three distinct regions between GBM, LGG, and normal tissue, with some overlap of LGGs in the GBM and normal tissue regions. A selected 35 of the 438 total patient data files were used to ensure balanced representation between all patient data and the GBO data. This selection aimed for 6–11 patient files for the following subcategories: normal, LGG‐normal, LGG, LGG–GBM, and GBM based on total PCA results. In short, six patient data with the highest variance between normal and GBM sets were selected. The LGG patient data points that closely overlapped with the selected normal and GBM patients were labeled as LGG‐normal and LGG–GBM, respectively. The six median LGG patients between LGG‐normal LGG–GBM were labeled LGG.

4.4. RNA sequencing

JX6 spheroids and GBOs were flash frozen without dissociation before RNA extraction. RNA was extracted according to the manufacturer's protocol for tissue samples (GeneJet RNA Purification Kit, K0731) and sequenced in Illumina NovaSeq 6000 for 25 M paired reads at HudsonAlpha's Discovery Life Sciences. FASTQ and BAM files were generated along with gene and transcript count analyses. A counts matrix was generated in‐house in R (v4.5.1) to perform differential gene expression analyses using the DESeq2 R Package (v1.48.2) following the standard workflow as previously published. ^{79 , 80}

4.5. Transcriptomics analyses and visualizations

The GBO counts matrix was generated to GRCh38.p13 to match patient STAR—Counts in‐house in R (v4.5.1). Differential gene expression analyses using the DESeq2 R Package (v1.48.2) following the standard workflow as previously published. ^{79 , 80} Principal component analysis (PCA) was performed after performing variance stabilizing transformation on the DESeq data and visualized using ggplot2 R Package (v4.0.0). A Venn Diagram was created with ggvenn R Package (v0.1.19) by extracting significant differentially expressed genes (DEGs) with p‐adjusted <0.05 for LGG versus normal and GBM versus normal. Volcano plots were performed and visualized using the EnhancedVolcano R Package (1.26.0) with a 0.05 p‐adjusted cut off, and 1.5‐fold change (FC) cut off. Heatmaps were generated using the ComplexHeatmap R Package (v2.24.1) filtering genes with p‐adjusted <0.05 and ordering the highest 50 log2FC for upregulated genes, and lowest 50 log2FC for downregulated genes. Gene ontology analyses were performed using the clusterProfiler R Package (v4.16.0) as previously published. ^{81 , 82}

Trajectory line plots were generated using an in‐house database for genes related to NVU zonation, cells, and basement membrane components. All patient and GBO data were combined for normalization to counts per million (CPM) to account for library size differences. Normal patient data and spheroids were chosen as references for patient and GBO data, respectively. The reference mean was calculated by averaging the CPM of each gene across the references. The relative expression was determined by dividing the CPM of every patient and GBO sample by the respective reference mean. A log2 transformation of the relative expression ratios was performed to ensure symmetry. The database for the trajectory line plots and their corresponding measurements is described in Table 1.

TABLE 1.

Gene database for neurovascular (NVU) zones, cell types, and basement membrane components.

Category	Subcategory	Gene symbol	Ref.
NVU zones	Arterioles	AIF1L	17, 22
		CD320	17, 22
		LGALS3	17, 22
		VEGFC	22, 88
		VSIR	17, 22
	Capillaries	ANGPT2	17, 22
		ANO2	17, 22
		CA2	17, 22
		CD74	17
		ESM1	17, 22
		HLA‐DRA	17
		MFSD2A	17, 22
		PLVAP	17, 22
		SLC38A5	17, 22
		SLC7A5	17, 22
		TFRC	17, 22
	Large arteries & arteries	ADAMTS1	17, 22
		BMX	17, 22
		EFNB2	17, 22
		FBLN5	17, 22
		GJA5	17, 22
		HEY1	17, 22
		IGFBP3	17, 22
		LTBP4	17, 22
		SEMA3G	17, 22
	Large veins/veins	CCL2	17, 22
		DNASE1L3	17, 22
		POSTN	17, 22
		PTGDS	17, 22
		SELE	17, 22
		SELP	17, 22
	Venules	ADGRG6	17, 22
		BNC2	17, 22
		JAM2	17, 22
		NR2F2	17, 22
		PRCP	17, 22
		PRSS23	17, 22
		RAMP3	17, 22
		TSHZ2	17, 22
Cell	Astrocyte	ALDH1L1	89
		AQP4	89
		GFAP	22, 89, 90
		S100B	89
		SOX9	89
	Endothelial	CD34	91
		CDH5	89
		CLDN5	89
		ENG	92
		KDR	93, 94
		PECAM1	17, 22, 89
		PROM1	91
		TJP1	89
		VWF	17, 92, 93
	Fibroblast	CEMIP	22
		FAP	92
		FBLN1	22
		KCNMA1	22
	Pericyte/smooth muscle	ACTA2	22, 89, 95
		CSPG4	89
		GRM8	22
		PDGFRB	22, 89
		SLC20A2	17, 22
		SLC30A10	17, 22
Basement membrane/extracellular matrix (ECM)	Adhesion	CLDN5	17, 89, 90
		F11R	96
		JAM2	96
		JAM3	96
		OCLN	97
		TJP1	89, 90
		TLN1	98
		TLN2	98
	ECM—fibrous	COL18A1	99
		COL1A1	92
		COL1A2	92
		COL4A1	17, 100
		COL4A2	17, 100
		FN1	92
		ITGA7	90
		LAMA2	90, 100
		LAMA4	90, 100
		LAMA5	90, 100
		TNC	92
	ECM—proteoglycan	AGRN	99, 100
		BCAN	99, 101
		CSPG5	99
		EGFLAM	99
		ESM1	17
		EYS	99
		GPC1	99
		GPC2	99
		GPC3	99
		GPC4	99
		GPC5	99
		GPC6	99
	ECM—proteoglycan	HSPG2	99, 100
		NCAN	99, 101
		NRXN1	99
		PTPRZ1	99, 101
		PTPRZ1	99
		SDC1	99
		SDC2	99
		SDC3	99
		SDC4	99
		SPOCK1	17, 99
		SPOCK2	17, 99
		SPOCK3	17, 99
		SRGN	99

4.6. Fluorescent staining

Frozen sectioning was performed as previously published. ⁴⁵ Whole mount spheroids were stained in mini centrifuge tubes, pipetted onto a cavity slide (BI0086B, Eisco™) with a cut 100 μL pipette tip, then mounted with ProLong™ Diamond Antifade Mountant (P36965, Invitrogen™) with a 12 mm circular coverslip (89015‐725, VWR). Primary antibody and secondary staining were performed according to Table 2. Cell nuclear staining was performed with Hoechst 33342 (H1399, Invitrogen™). Fluorescent confocal imaging was performed using the Nikon C2 Laser Scanning Microscope.

TABLE 2.

Primary antibody (1° AB) and secondary (2° AB) staining chart.

1° AB	1° AB supplier	1° AB host species	1° AB dilution ratio	2° AB	2° AB dilution ratio	2° AB supplier
Tenascin‐C	ab108930, Abcam	Rb a	1:500	Gt b anti‐Rb488	1:1000	A11008, Invitrogen™
Fibronectin	610,077, BD Biosciences	Ms c	1:500	Gt anti‐Ms555	1:1000	A21422, Invitrogen™
CD31	ab56299, Abcam	Rt d	1:50	Gt anti‐Rt647	1:1000	A21247, Invitrogen™

^a Rb: rabbit.

^b Gt: goat.

^c Ms: mouse.

^d Rt: rat.

4.7. Morphometric analyses

Vascular‐like morphometrics were quantified using ImageJ (v1.54g) via a custom macro. For each CD31+ stained GBO (n = 4), three concentric regions of interest (ROIs) were defined based on spatial protein expression: rim, vascular‐like area (VLA), and core. Images were converted to 8‐bit, and a Li Threshold Mask was applied to isolate CD31+ stains from the background. Lumens were defined as enclosed, non‐fluorescent voids within a continuous CD31+ boundary. Vessel junctions were identified as points of intersection between three or more CD31+ branches and quantified as per unit area (mm²). The distance between neighboring lumens was measured using the centroid‐to‐centroid method. The minimum ferret diameter was utilized to quantify lumen scale, with distributions calculated from the representative image in Figure 4a. To account for inter‐sample variability in total vascular density, junction density, and lumen spacing were normalized to a fractional value relative to the internal sample mean. Regional trends in fractional junction density and lumen spacing were evaluated using a one‐way repeated measures ANOVA to account for the paired nature of the ROIs within each biological replicate. Significance was defined as p < 0.05.

4.8. FITC‐dextran diffusion and multiphoton image analysis

Diffusion throughout the 2 mm GBOs and through self‐assembled vascular‐like structures was determined using dextran, fluorescein, and biotin, 3000 MW, Anionic, Lysine Fixable (Micro‐Emerald) (D7156, Invitrogen™). GBOs were collected and incubated in FITC‐dextran for 0, 1, and 6 h (n = 6 biological replicates) in 5% CO₂ at 37°C then gently washed in 1× PBS for 5 min. Next, the GBOs were counterstained with 1:1000 Hoechst in 0.1% Triton X‐100 in PBS solution for 30 min, washed with 1× PBS for 5 min again before being fixed in 4%‐paraformaldehyde at 4°C overnight. The fixed GBOs underwent a thorough washing step with 1× PBS for 5 min three times. The size and density of GBOs required clearing with RapiClear (RC152001, SUNJin Lab) overnight at room temperature then stored at 4°C until imaging.

FITC‐dextran and Hoechst counterstained GBOs were imaged at 800 nm on the Nikon A1R MP+ multiphoton microscope (Figure 5a). The whole surface area of the GBO was imaged ~800–850 μm deep into the GBO with a 0.5 μm resolution. Because each image was >60 GB, image analysis was performed on The University of Alabama's High Power Computer using the following Python packages: nd2, dask, scipy.ndimage, skimage.morphology, and matplotlib. Portions of the Python analysis pipeline were developed with assistance from Google Gemini (Gemini 1.5 Pro, accessed June 2026). Gemini was used for optimizing memory‐mapped ND2 file reading, debugging high‐performance Dask array slicing routines on the HPC cluster, and refactoring the intensity‐weighted centerline perfusion plotting scripts. All code was reviewed, verified on independent tissue datasets, and modified by the authors, who take full responsibility for correctness and reproducibility.

To determine the diffusion coefficient and Fick's diffusivity, 3 GBOs at 0, 1, and 6 h were imaged where the 35th, 50th, and 65th z‐slices out of 100 were analyzed for intensity profiles across the whole GBO (Figure S2). The diffusion coefficient was solved using Equation 1:

$$\frac{\partial C}{\partial t}={\mathcal{D}}_{\mathrm{eff}}\left(\frac{{\partial }^{2}C}{d{r}^2}+\frac{2}{r}\frac{\partial C}{\partial r}\right)$$ (1)

where $C$ represents the net normal FITC‐dextran fluorescence intensity, $t$ is time, $r$ is radial distance in μm, and ${\mathcal{D}}_{\mathrm{eff}}$ is the localized effective diffusion coefficient in μm²⋅s⁻¹. The Fickian model was determined using Equation 2:

$$\frac{\partial C\left(r,t\right)-C_0}{C_1-C_0}=1+\frac{2R}{\mathit{\pi r}}\sum _{n=1}^{\infty }\frac{{\left(-1\right)}^n}{n}\sin \left(\frac{\mathit{n\pi r}}{R}\right)\exp \left(-\frac{{\mathcal{D}}_{\mathrm{eff}}{n}^2{\pi }^{2}t}{R^2}\right)$$ (2)

where $R$ is the maximum boundary radius of the GBO, $C_0$ is the initial baseline core fluorescence intensity at $t=0$, and $C_1$ represents the peak source fluorescence intensity stabilized at the peripheral infiltration boundary layer (Figure 5b).

To analyze channel geometry and tracer diffusion within the 2 mm GBO NVU‐like regions, high‐resolution region of interest (ROI)‐crops were extracted from the 6 h maximum intensity projections (Figure 5ci). Foreground networks were isolated by converting frames to 8‐bit matrices, applying a localized adaptive thresholding mask, and removing disconnected background noise particles (<50 px). To visually validate transit along continuous tracks, the binarized mask was skeletonized to a 1 px wide centerline map (Figure 5cii). ^{83 , 84 , 85} Local fluorescence intensity was sampled along the skeleton by taking the mean pixel values within a 3 × 3 px neighborhood to suppress point noise. These intensity values were mapped back onto the centerline using a thermal scale and overlaid on a dim binary foreground canvas to illustrate active diffusion pathways. ^{86 , 87} Vessel calibers were profiled by applying a 2D Euclidean Distance Transform (EDT) to the foreground mask to calculate the distance from interior channel pixels to the nearest background edge. Local vessel diameters along the network centerlines were computed as twice the local radius and calibrated to μm using objective metadata. The diameter heatmap range was bounded between 0 and 10 μm to maximize visual dynamic contrast and highlight physiological capillary‐scale lumen caliber transitions across the NVU‐like regions.

4.9. Statistical analysis

Data visualization and statistical analyses were performed in R (v4.5.1) or GraphPad Prism (v9.0) as described in Patient Data Retrieval from TCGA, Transcriptomics via RNA‐seq, and Fluorescent Staining sections. In short, DESeq2 was utilized to evaluate statistical significance of differential expression for all RNA‐seq related work. Fluorescent staining image analysis was performed on n = 4 and statistically analyzed in GraphPad Prism. Prior to statistical testing, datasets were screened for outliers using the Grubbs' test. Two outliers were identified and removed from the lumen spacing dataset (one each in the core and vascular‐like regions) to ensure normality and variance homogeneity. Statistical significance for the remaining fractional datasets was then evaluated using a one‐way ANOVA with a Tukey–Kramer HSD post hoc test. All error bars represent the standard error about the mean.

5. CONCLUSIONS

In summary, this study demonstrates that GBOs successfully recapitulate a conserved clinical glioma signature, providing the essential biochemical cues and biophysical architecture necessary for autonomous NVU‐like assembly. While our model achieves high morphological correspondence with patient‐derived capillary scales, the observed transcriptomic divergences in arterial, endothelial, and proteoglycan markers suggest that the GBOs have limited internal hemodynamic shear stress (HSS) for full physiological maturation. To bridge this functional gap, we propose integrating GBOs into a perfusion bioreactor system capable of maintaining the bulk shear stress of our current system (~0.5 Pa) to promote peripheral expansion while simultaneously introducing internal HSS to stabilize endothelial cell networks. We suggest that this perfusion be initiated at the 1–1.5 mm developmental stage, specifically to allow for the critical, self‐organized fabrication of the TNC and FN matrix identified as a prerequisite for vascular recruitment. Ultimately, these GBOs serve as a high‐fidelity, high‐throughput modeling system for glioblastoma development and a specialized platform for the validation of ECM‐targeted therapies. By targeting the TNC/FN scaffold, such therapies may weaken the pathological barrier of the tumor niche, potentially facilitating enhanced drug diffusion across the NVU and improving therapeutic outcomes in GBM.

AUTHOR CONTRIBUTIONS

Alexandra D. Avera: Conceptualization; investigation; methodology; software; data curation; validation; formal analysis; visualization; writing – original draft; writing – review and editing. Taylor N. Schnorbus: Investigation; validation; writing – review and editing; writing – original draft. Yonghyun Kim: Conceptualization; methodology; writing – review and editing; project administration; supervision; resources; visualization; funding acquisition; formal analysis.

FUNDING INFORMATION

This work was supported by the National Science Foundation (CBET/EBMS #2000053 to Y.K.) and by The University of Alabama Division of Research (A26‐0236‐001 to Y.K.). A.D.A. was supported by the U.S. Department of Education as a GAANN Fellow (P200A210069; PI: Y.K.). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funding institutions.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

Figure S1. GSC and ECM components TNC and FN characterization. Heatmaps to characterize the heterogeneity of JX6 source cells (single_cell) as GBOs grow larger (a) compared to TCGA‐derived patient data (b). These heatmaps were specifically analyzed for GSC markers SOX2, OLIG2, NES, MSI1, POU5F1, PROM1, and ECM component markers TNC and FN1.

Figure S2: Image processing of FITC‐dextran transport kinetics in 2 mm GBOs. (a‐i) Representative maximum intensive projection of 6 h stained GBO volume where a 50 μm z‐stack with 0.5 μm step resolution across the core of the GBO was analyzed. (a‐ii) Z‐stacks 35, 50, and 65 were chosen to analyze (a‐iii) 10 intensity profile lines that are 18° apart for even distribution around the GBO. (b) Mean intensity profile plots with respect to radial distance where the center is r = 0 μm across all three time points (0, 1, and 6 h) and all three z‐slices (z = 35, 50, 65). (c) Overlay of macro‐diffusion timeline of FITC‐dextran of across all time courses and z‐slices. Data represents n = 3 biological replicates per time course with three technical replicates (z‐stacks); 95% confidence intervals.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.