Engineering spatially defined extracellular matrix gradients to govern self-organized multicellular aggregates in a glioblastoma-on-a-chip

Abstract

The spatial heterogeneity of biochemical cues within the tumor microenvironment (TME) critically influences cancer progression and therapeutic resistance. However, existing models often lack the capacity to generate stable, quantitative concentration gradients in a high-throughput and biomimetic manner. Here, we present a glioblastoma-on-a-chip platform featuring a 48-microwell array that enables spontaneous formation of spatially defined extracellular matrix (ECM) gradients through a structure-guided solution replacement process. This integrated strategy combines gradient generation, sample arraying, and gel solidification into a single step, allowing one-step fabrication of gelatin methacryloyl (GelMA) microgel arrays with 48 discrete concentration conditions. We successfully generated continuous fibronectin (FN) gradients, quantitatively validated with a linear standard curve (R² = 0.9899) and categorized into five statistically distinct groups (**p < 0.01). Computational fluid dynamics simulations confirmed physiological flow perfusion in the microchannels, providing essential biophysical TME cues. When applied to 3D dynamic co-culture of U87 glioblastoma and vascular endothelial cells (HUVECs), the FN gradient critically regulated the formation of self-organized multicellular aggregates, showing strong concentration dependence in their probability, number, and size. These aggregates exhibited significant upregulation of cancer stem cell markers (CD133, Vimentin, α-SMA), with CD133 expression increased by over 559-fold compared to the control group. Compared to conventional 96-well plate MTT assays, the multicellular aggregates demonstrated enhanced resistance to temozolomide, highlighting its utility for drug response studies in a physiologically relevant context. This work establishes a robust platform for constructing quantitative ECM gradients and serves as a potential tool for investigating cell–ECM interactions and high-throughput drug screening within biomimetic TMEs.

---

1. Introduction

The progression and therapeutic resistance of solid tumors, including glioblastoma (GBM), are not solely dictated by the genetic alterations within cancer cells but are profoundly influenced by their dynamic interactions with the tumor microenvironment (TME).^1–3 The TME is a complex and adaptive ecosystem composed of diverse cellular components, such as endothelial cells and fibroblasts, and a unique non-cellular milieu—the extracellular matrix (ECM).⁴ Beyond providing structural scaffolding, the ECM serves as a critical signaling hub, presenting a spatially heterogeneous landscape of biochemical and biophysical cues.^5,6 Gradients of ECM proteins, such as fibronectin (FN), laminin, and collagens, are not static backdrops but are actively remodeled during tumor invasion and metastasis.⁷ In GBM, this spatially defined ECM topography is increasingly recognized as a key regulator of malignant phenotypes.⁸ It guides the invasive behavior of tumor cells along specific pathways, helps maintain pools of therapy-resistant cancer stem cells (CSCs) within protective niches, and modulates the delivery and efficacy of chemotherapeutic agents.⁹ The ability of neoplastic and stromal cells to sense, remodel, and respond to these graded ECM signals is therefore a fundamental driver of tumor progression and a significant contributor to poor clinical outcomes.^10,11

The critical need to recapitulate this in vivo context has driven the evolution of in vitro cancer models.¹² While indispensable for basic research, conventional two-dimensional (2D) cell cultures on rigid plastic substrates present a highly simplified and physiologically inaccurate environment.^13,14 They fail to capture the three-dimensional (3D) architecture that governs cell–ECM interactions, force transduction, and paracrine signaling, leading to a well-documented disparity in gene expression, drug sensitivity, and cellular behavior compared to in vivo conditions.¹⁵ The advent of 3D hydrogel systems, employing natural or synthetic polymers such as gelatin methacryloyl (GelMA), has marked a significant advancement by providing a biomimetic scaffold that allows for more realistic cell morphology and microenvironmental sensing.^16–19 However, a pervasive limitation of most standard 3D models is their inherent homogeneity; they typically present biochemical cues in a uniform manner, thereby failing to capture the spatially graded concentrations of ECM components and soluble factors that are a hallmark of native tissue and tumor stroma.^20,21 This inability to engineer controlled spatial heterogeneities in vitro limits our understanding of how cells interpret and navigate complex biochemical landscapes, a process central to cancer metastasis and treatment failure.

To bridge this gap, organ-on-a-chip (OoC) technologies have emerged as a powerful tool, offering precise control over the cellular microenvironment to recapitulate tissue- and organ-level physiology.^22–25 However, a significant challenge within the OoC field remains the generation of stable, long-range biochemical gradients that seamlessly balance operational simplicity, high throughput, and precise quantitative control.^26–28 Existing methods, such as pump-driven fluidic networks, can generate physiologically relevant shear stresses. However, they are often bulky, require expert operation, and can subject cells to non-physiological forces. In contrast, simpler, diffusion-based alternatives typically suffer from limited spatial range and transient stability.²⁹ This fundamental trade-off, coupled with multi-step procedures that risk compromising cell viability and hinder scalability, has ultimately constrained the widespread adoption of robust gradient-based models for high-throughput applications like drug screening.

Herein, we report a novel microfluidic platform that leverages intrinsic chip microstructures to generate wide-range, quantitative, and uniform biochemical gradients within a biocompatible GelMA hydrogel in a single step, without complex external drives. This integrated system seamlessly combines gradient generation, 3D cell encapsulation, and high-throughput culture. As a proof of concept, we engineered a glioblastoma-on-a-chip featuring a high-density microwell array to establish precise FN gradients. We demonstrate that these gradients govern the self-organization of GBM and endothelial cells into multicellular aggregates and modulate functional outcomes, including stemness marker expression and chemoresistance. This work advances GBM models by providing a convenient and high-throughput platform for gradient-based studies within biomimetic TMEs.

2. Materials and methods

2.1 Materials

Gelatin methacryloyl (GelMA), 2-hydroxy-4′-(2-hydroxyethoxy)-2-methylpropiophenone (I2959), and FITC-labelled dextran (FITC-dextran, 20 kDa) were purchased from Shanghai Aladdin Biochemical Technology Co., Ltd.; dimethyl sulfoxide (DMSO), 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI) were purchased from Sigma; recombinant human fibronectin (FN), blocking buffer, antibody diluent, Alexa Fluor™ 647 anti-CD133 antibody, rabbit anti-CD31 antibody, rabbit anti-α-smooth muscle actin (α-SMA) antibody, rabbit anti-vimentin (VIM) antibody, and goat anti-rabbit IgG568 were purchased from Abcam; Cell Tracker™ Red CMFDA, LIVE/DEAD™ Cell Imaging kit, PureLink™ RNA Mini Kit, propidium iodide (PI), Alexa Fluor®488 phalloidin, and TRIzol® reagent were purchased from Life Invitrogen; Dulbecco's modified Eagle medium with 4.5 g L⁻¹ glucose (HG-DMEM), penicillin–streptomycin double antibody, 0.25% trypsin–EDTA, and fetal bovine serum (FBS) were purchased from Gibco; endothelial cell medium was purchased from ScienCell; One Step RT-qPCR Kit (dye method), phosphate buffered saline (PBS) tablets, and temozolomide were purchased from Sangon Biotech (Shanghai) Co., Ltd.; acetic acid (glacial acetic acid) was purchased from Sinopharm Chemical Reagent Co., Ltd.; collagen type I (rat tail) was purchased from Corning; MTT assay kit was purchased from Solarbio; polydimethylsiloxane (PDMS, Sylgard 184) kit (including initiator) was purchased from Dow Corning. All reagents were strictly stored and used according to experimental requirements.

2.2 Chip design and fabrication

In this study, PDMS microfluidic chips were fabricated using soft lithography following standard protocols.³⁰ The chip consisted of one liquid inlet, one liquid outlet, and 48 structurally identical microwells, with microchannel height and width of 225 μm and 200 μm respectively. The microwell diameter was 900 μm, with a front gap width of 200 μm, narrow gap width of 40 μm, and length of 140 μm. The chip design and images are shown in Fig. 1-A.

Fig. 1 Chip design and schematic diagram of concentration gradient formation. (A) Photograph of the microchip. Scale bar = 5 mm. The illustration was a photo of the microwell by microscopy. Scale bar = 100 μm. (B) Schematic diagram of the concentration gradient formation.

2.3 Preparation and characterization of microgels

This study used an aqueous solution containing 5% (w/v) GelMA and 0.5% (w/v) photoinitiator I2959 as the prepolymer solution for preparing GelMA microgels. The specific preparation process was as follows: first, the prepolymer solution was injected into the microchip through Tygon tubing connected to a syringe pump (Harvard Apparatus) at a flow rate of 8 μL min⁻¹ until the microchannels and microwells were completely filled; subsequently, air was introduced at a flow rate of 30 μL min⁻¹ from the outlet to reverse-flush the solution in the microchannels; then, the prepolymer solution mixed with additives was injected at a flow rate of 2 μL min⁻¹, and the injection was stopped after the solution passed through the last microwell (48#); after standing for 30 min; finally, air was again introduced at a flow rate of 30 μL min⁻¹ to expel the residual solution in the microchannels. At this point, the remaining prepolymer solution mixed with additives in the microwells was cured by 365 nm UV irradiation for 45 s, forming GelMA microgels with additive concentration gradients.

The pure GelMA microgels (without any additives) were characterized by Fourier transform infrared spectroscopy (FTIR) (Nicolet iS50, Thermo Scientific). When the additives were FITC-dextran and FN respectively, their final concentrations in the GelMA prepolymer solution were 2 mg mL⁻¹ and 200 ng mL⁻¹. The obtained FITC-dextran concentration gradient GelMA microgels could be quantitatively characterized by fluorescence using an inverted fluorescence microscope (IX-73, Olympus). Meanwhile, FITC-dextran GelMA microgels with concentrations of 0.02 mg mL⁻¹, 0.125 mg mL⁻¹, 0.25 mg mL⁻¹, 0.5 mg mL⁻¹, 0.75 mg mL⁻¹, 1 mg mL⁻¹, 1.5 mg mL⁻¹, and 2 mg mL⁻¹ could be formed in the chip microwells by direct addition method. The fluorescence intensity was measured separately to establish a standard curve.

2.4 Cell culture

This study used green fluorescent protein-transfected human brain astrocytoma cells (U87-eGFP) and human umbilical vein endothelial cells (HUVECs) for experiments. U87-eGFP cells were purchased from iCell Bioscience Inc. (Shanghai, China), while HUVECs were purchased from the Cell Bank/Stem Cell Bank of the Chinese Academy of Sciences (ATCC Source). U87-eGFP cells were cultured in HG-DMEM containing 10% (v/v) FBS and 1% (v/v) penicillin–streptomycin double antibody. HUVECs were cultured in specialized endothelial cell medium, and the culture flasks required collagen coating before use: first, a 3% (v/v) Collagen Type I working solution was prepared using filter-sterilized 0.1% (v/v) glacial acetic acid solution, evenly applied to the flask surface with a sterile rubber-tipped pipette. After standing, excess liquid was removed, and the flasks were dried in a 50 °C oven for 4 h to remove residual acetic acid. All cells were routinely cultured in T25 flasks under conditions of 37 °C, and 5% CO₂. When reaching 75–85% confluency, cells were passaged using 0.25% trypsin–EDTA, with all operations strictly following aseptic protocols.

2.5 Cell seeding and 3D culture in microchip

This study utilized 0.25% trypsin–EDTA to digest U87-eGFP cells or HUVECs. In the monoculture study, cells were resuspended in GelMA prepolymer solution at a density of 2 × 10⁶ mL⁻¹; in the coculture study, HUVECs and U87-eGFP cells were mixed at a 2 : 1 cell ratio before resuspension in GelMA prepolymer solution. When constructing the concentration gradient microenvironment, the cell/GelMA mixture served as the prepolymer solution, while the FN-supplemented GelMA mixture acted as the additive solution. Cell-laden microgel arrays were prepared using the aforementioned method for subsequent experiments.

2.6 Fluid perfusion

Based on hydrodynamic simulations, a model was established according to the chip microchannels to simulate flow rates for dynamic cell culture, ensuring alignment with physiological flow velocities. The optimized dynamic culture condition was 0.1 μL min⁻¹, which was uniformly applied in all perfusion operations via syringe pump from the chip inlet.

2.7 Cell viability assay

During dynamic cell culture, cell viability and proliferation were characterized using live/dead staining and MTT assays. For HUVECs, the calcein-AM/PI viability kit was employed per manufacturer instructions; for U87-eGFP cells, intrinsic green fluorescent protein and PI staining were used. All images were acquired via confocal microscopy (FV 3000, Olympus). MTT assays were performed on days 0, 3, and 5 to assess proliferation of U87-eGFP cells and HUVECs in 3D microwells arrays under dynamic conditions. The procedure was performed as follows: the supernatant was aspirated, and 100 μL of a mixture consisting of 90 μL fresh medium and 10 μL MTT solution was added to each sample. After incubation at 37 °C for 4 hours, the supernatant was carefully removed. The formed formazan crystals were then dissolved in 110 μL of formazan solution. The resulting solution was transferred to a 96-well plate, and its absorbance was measured at 490 nm using a full-wavelength microplate reader (Multiskan SkyHigh, Thermo Scientific).

2.8 Live cell tracing

To distinguish cell types in coculture, HUVECs were specifically labeled with Cell Tracker™ Red CMFDA according to the manufacturer's instructions, enabling panoramic fluorescence imaging of U87-eGFP cells (green) and HUVECs (red) under inverted fluorescence microscopy for real-time monitoring within the concentration gradient ECM.

2.9 F-actin staining and immunofluorescence staining

Characterize the cytoskeletal actin (F-actin) of HUVECs within microwells using Alexa Fluor® 488 phalloidin. Fluorescent images of cells were acquired at two culture time points (18 h, 3 days) using an inverted fluorescence microscope.

Detect the expression of α-SMA, VIM, CD133, and CD31 via immunofluorescence staining. Methodology: first, fixed with 4% paraformaldehyde at room temperature for 40 min; treated with 1 : 10 diluted blocking agent for 2 h; then added 1 : 100 diluted primary antibodies and incubated overnight at 4 °C (for CD133, directly used Alexa Fluor™ 647-labeled primary antibody, omitting the secondary antibody step); the next day, incubated with 1 : 100 diluted Alexa Fluor™ 568-labeled secondary antibody at room temperature for 2 h, and the cell nuclei were stained with DAPI for 10 min. Between steps, washed 2–3 times with PBS (each for 1 min). Finally, preserved samples in PBS and performed confocal laser microscopy imaging. This protocol accounted for 3D culture characteristics by appropriately extending processing times to ensure full penetration of staining reagents into the gel.

2.10 Drug testing

In the chip-based 3D dynamic culture, all groups underwent a standardized 3 day pre-culture period to form aggregates. Four experimental conditions were then evaluated from day 4 to day 5 (a 48 hour treatment period): 1) U87-eGFP culture in plain GelMA (without FN); 2) U87-eGFP culture in plain GelMA with 150 μM TMZ; 3) U87-eGFP/HUVECs co-culture in FN-containing (50 ng mL⁻¹) GelMA; 4) U87-eGFP/HUVECs co-culture in FN-containing GelMA with 150 μM TMZ. This design ensured an identical total culture duration (5 days) for valid resistance assessment.

2.11 Total RNA isolate and RT-qPCR

This study constructed uniform GelMA microgels with a final FN concentration of 50 ng mL⁻¹ in microwells. After 5 days of perfusion with culture medium, total RNA was first extracted using TRIzol® reagent, then isolated following the protocol of the PureLink™ RNA Mini Kit. Subsequently, cDNA synthesis and PCR reactions were performed on the total RNA according to the one-step RT-qPCR kit instructions, with detection carried out using the Lightcycler® 96 (Roche). The expression level of each gene was normalized to β-actin as an internal reference. Investigated genes included α-SMA, VIM, and CD133, with primer sequences as follows:³¹

α-SMA: 5′-AAAAGACAGCTACGTGGGTGA-3′ (forward), 5′-GCCATGTT-CTATCGGGTACTTC-3′ (reverse);

VIM: 5′-GACGCCATCAACACCGAGTT-3′ (forward), 5′-CTTTGTCG-TTGGTTAGCTGGT-3′ (reverse);

CD133: 5′-TTCTTGACCGACTGAGACCCA-3′ (forward), 5′-TCATGTTC-TCCAACGCCTCTT-3′ (reverse);

β-actin: 5′-CTGTCTGGCGGCACCACCAT-3′ (forward), 5′-GCAACTAA-GTCATAGTCCGC-3′ (reverse).

2.12 Statistical analysis

This study employed multiple analytical tools for data processing: PCR results were analysed using Lightcycler® 96 software; image processing, including multi-field image stitching, fluorescence intensity quantification, cell cluster counting, and cluster size measurement, was performed with Image J software, inverted fluorescence microscopy, and confocal microscope built-in software. Unless otherwise specified, all experimental data were presented as mean ± standard deviation (mean ± SD), with each experiment repeated at least 3 times (n ≥ 3). Statistical significance was determined by Student's t-test when *p < 0.05, **p < 0.01 and ***p < 0.001.

3. Results and discussion

3.1 Preparation and characterization of concentration gradient microgels

In this study, we established an innovative method for achieving concentration gradients through precise fluid control based on a specific PDMS microfluidic chip. Systematic fluorescence imaging and statistical analysis confirmed that the gradient exhibited a wide range, high continuity, and statistically significant concentration differences. This chip, as shown in Fig. 1-A, enabled convenient preparation of various microgels with uniform or biochemical molecular concentration gradients obtained through UV curing (e.g., GelMA). Water-soluble photoinitiators such as I2959 were used, while the biochemical molecules encompassed various dyes, small drug molecules, and proteins, as illustrated in Fig. 1-B. For constructing TME or tumor-on-a-chip models, we employed GelMA hydrogel due to its excellent biocompatibility, making it particularly suitable as a 3D culture scaffold for vascular endothelial cells and tumor microtissues. This facilitated our investigation of the interactions between tumor cells and vascular endothelial cells during tumorigenesis, progression, and metastasis. We successfully prepared GelMA microgels with varying concentrations of FITC-dextran (20 kDa) additives. The schematic diagram and process microscopy images of the specific operations are shown in Fig. 2-A and B. The realization of the concentration gradient is based on a PDMS microfluidic chip with a specific structure and is completed through a set of precise fluid operation procedures: (1) the GelMA prepolymer aqueous solution containing the photoinitiator was injected into the microchannel from the inlet to fill the entire microchip. (2) Air was rapidly introduced from the outlet in reverse to expel the solution from the microchannel. Due to the microchannel's structure, the solution in the microwells was constrained and not expelled. If UV polymerization was performed directly at this stage, uniform GelMA microgels were obtained. In order to obtain microgels with concentration gradients, the following steps were necessary. (3) The GelMA prepolymer solution mixed with the FITC-dextran additive was injected into the microchannel again from the inlet, stopping immediately after the solution flowed through the last microwell (48#). The injection volume was experimentally calculated and optimized to ensure the solution front reached the 48th microwell, with the optimal volume being approximately 30 μL in this study. This operation caused the newly injected GelMA solution containing FITC-dextran to undergo material exchange with the initially stored GelMA prepolymer solution in each microwell as it flowed through. The solution continuously lost FITC-dextran molecules and became diluted, while the original GelMA prepolymer solution in the microwells mixed with varying amounts of FITC-dextran molecules, forming a concentration gradient. (4) The system was allowed to stand for 30 min to stabilize the molecule exchange in the microwell, then air was rapidly introduced in reverse again to expel the liquid in the channel. After UV photopolymerization, microgels with an FITC-dextran concentration gradient were obtained.

Fig. 2 Synthesis and characterization of GelMA microgels with FITC-dextran concentration gradient on a microchip. (A) Schematic diagram of the operations. (B) Bright field and fluorescent images of the specific processes. Scale bar = 1 mm. (C) Statistical analysis of relative fluorescence intensity of different microwells. Microwells were numbered in the order of fluid direction from 1# to 48#. **p < 0.01. The panoramic fluorescence image was the GelMA microgels with FITC-dextran concentration gradient on a chip, which was stitched from four fluorescence photographs taken under identical imaging conditions using a 1.25× objective lens. Scale bar = 1 mm.

The verification of the concentration gradient is mainly carried out through fluorescence imaging and quantitative statistical analysis. For the obtained homogeneous GelMA microgels, we characterized their composition using FTIR spectroscopy, and the results are shown in Fig. S1. Comparative spectral analysis revealed characteristic peaks of GelMA appearing around wavenumber 1550 cm⁻¹, associated with N–H bending vibration and C–N stretching vibration.³² The FTIR results demonstrated that the composition of GelMA microgels prepared by the chip method was identical to those prepared by non-chip methods. Additionally, for the GelMA microgels with FITC-dextran concentration gradients generated on chip, fluorescence imaging was employed to characterize the concentration gradients, as illustrated in Fig. 2-C. The panoramic fluorescence image was stitched from four fluorescence photographs taken under identical imaging conditions using a 1.25× objective lens. From the 1# microwell closest to the inlet to the 48# microwell, the fluorescence intensity gradually decreased, indicating a progressive reduction in the concentration of FITC-dextran molecules incorporated into the microwells. We quantitatively characterized the relative fluorescence intensity of concentration-gradient microgels obtained from multiple chips, observing a continuous decrease in relative fluorescence intensity with increasing microwell numbers. Subsequently, we employed the t-test to group the microwells based on fluorescence intensity. When the relative fluorescence intensity of subsequent microwells showed a statistically significant difference (**p < 0.01) compared to preceding microwells, they were categorized into distinct groups: G1 (highest concentration, 1#–13#), G2 (14#–27#), G3 (28#–36#), G4 (37#–43#), and G5 (lowest concentration, 44#–48#). Meanwhile, a standard curve for quantifying the gradient was established via the direct addition method, wherein FITC-dextran of specific concentrations was directly mixed into the GelMA prepolymer solution. This curve exhibited a linear relationship between concentration and fluorescence intensity (R² = 0.9899; Fig. S2), enabling the analysis of the actual FITC-dextran concentration in the gradient microgels and the estimation of the incorporation ratio for other biochemical molecules. The stability of the gradient prior to UV curing was confirmed by the highly consistent and continuous profile across multiple replicates (Fig. 2-C) and statistically significant inter-group differences (**p < 0.01), attesting to the robustness of the formation process. The 30 minute stabilization period was empirically determined to be sufficient for molecular diffusion and exchange to reach equilibrium.

This study departs from the traditional design that relies on complex external control. By utilizing the microwells structure of the microfluidic chip itself, the natural decreasing distribution of solute concentration is achieved by regulating the convection–diffusion and solution displacement that occur when the prepolymer solution containing additives flows through the microwells array filled with pure prepolymer solution. This method does not require external pumps, valves, or multi-layer chip structures, significantly reducing the system complexity and operating cost. Meanwhile, this scheme integrates gradient formation, sample arraying, and gel solidification on the same platform. After the gradient is stabilized, in situ polymerization is initiated by ultraviolet light to form a GelMA microgel array containing a preset concentration gradient in 48 microwells with the same structure at once, achieving “one-step” high-throughput preparation. Since all units are formed at one time in the same fluid environment, it ensures high uniformity and comparability of microgels at different concentration points and within groups, significantly reducing batch errors. Compared with the limitations of traditional methods with limited ranges and fixed morphologies, this scheme naturally forms a wide-range, continuously decreasing concentration gradient from the inlet to the outlet in a single run. Through statistical analysis, it can be divided into 5 concentration intervals with highly significant differences (**p < 0.01), demonstrating the high smoothness and distinguishability of the gradient. In addition, the standard curve of fluorescence intensity-concentration (R² = 0.9899) established by the “direct addition method” enables the accurate quantification and prediction of the actual molecular concentration in each micropore, providing a reliable quantitative basis for subsequent research.

3.2 3D culture of vascular endothelial cells and physiological fluid simulation

Based on the successful preparation of GelMA microgels, we conducted a 3D culture of vascular endothelial cells (HUVECs). The process was illustrated in Fig. 3-A. Initially, uniform HUVECs-loaded GelMA microgels were prepared to facilitate observation, with cells labeled using Cell Tracker™ Red CMFDA. Panoramic fluorescence imaging (Fig. 3-B) demonstrated effective cell loading into the microwells, with minimal cell residue in the microchannels and uniform cell density across microwells. To enable long-term culture under physiologically relevant conditions, we implemented a perfusion system. Computational fluid dynamics (CFD) simulation (Fig. 3-C) confirmed that at a perfusion rate of 0.1 μL min⁻¹, the fluid velocity in the microchannels was 0.7 × 10⁻³ m s⁻¹, matching the physiological range of capillary flow, while the significantly lower velocity within the microwells simulates interstitial flow without generating excessive shear stress. After dynamic culture, cell viability assessed by live/dead staining (calcein-AM/PI) revealed a survival rate exceeding 95% after 3 days (Fig. 3-D). The MTT assay (Fig. 3-E) showed a 1.57-fold and 2.36-fold increase in absorbance at 490 nm on days 3 and 5, respectively, compared to day 0, indicating robust proliferative activity.³³ The high cell viability and proliferation throughout the culture period confirm that this method, which is based on biocompatible GelMA and gentle fluid manipulation, achieves efficient 3D cell encapsulation without compromising cellular vitality. This makes it particularly suitable for constructing complex, living in vitro models such as the TME.

Fig. 3 Results of 3D cell culture in microwells and physiological fluid simulation. (A) Operation processes of preparation of HUVECs-loaded GelMA microgels with FN concentration gradient. (B) The panoramic fluorescence image of HUVECs-loaded GelMA microgels in a chip. HUVECs were labelled by Cell Tracker™ Red CMFDA. Scale bar = 1 mm. (C) Result of computational fluid simulation. (D) Results of cell viability and proliferation performed by calcein-AM/PI staining. Calcein-AM, green; PI, red. Scale bar = 200 μm. (E) MTT assay.

However, microscopic imaging revealed that despite high viability and proliferative capacity, HUVECs maintained a predominantly spherical morphology with poor spreading and pseudopod extension, potentially limiting intercellular interactions. To address this, we used FN as an additive, leveraging the chip's capability to easily generate biochemical molecular concentration gradients, thereby constructing FN-gradient GelMA hydrogels. Through F-actin cytoskeletal protein staining and CD31 immunofluorescence staining, we imaged HUVECs at different time points and under varying FN concentrations, as shown in Fig. 4. We compared the morphology and protein expression of HUVECs in the G1's 1# (the FN concentration was calculated to be approximately 61.2 ng mL⁻¹) and the G3's 28# (The FN concentration was calculated to be approximately 35.2 ng mL⁻¹) at 18 hours and 3 days post-seeding (Fig. 4-A). Notably, after 18 hours, HUVECs in 1# exhibited polygonal shapes with extended pseudopodia and clear CD31 expression. By day 3, the cells had fully spread into a cobblestone-like morphology, with CD31 localized at tight junctions between cells, forming a grid-like pattern. In contrast, HUVECs in 28# showed minimal morphological changes between 18 hours and 3 days. We further examined F-actin expression in other groups at day 3: 8# in the G1 group resembled 1#, while 18# in the G2 group displayed transitional states with some polygonal cells. The G4's 36# and G5's 46# were similar to 28# in the G3 group. These results suggested FN supplementation promoted endothelial cell spreading, potentially enhancing direct cell–cell contact and intercellular interaction capacity.³⁴ The successful incorporation of FN confirms that this method is applicable not only to model molecules like FITC-dextran but also to various bioactive factors including proteins and small-molecule drugs. This demonstrates broad material compatibility and application universality, providing a robust platform for accurately simulating in vivo biochemical microenvironments. It holds significant value for drug screening, tissue engineering, and disease model research.

Fig. 4 Fluorescence image of morphology and protein expression of HUVECs in GelMA microgels with different FN-gradient. (A) F-actin (green) and CD31 (red) expression of HUVECs at 18 h and 3 days in microwell 1# and 28#. (B) F-actin (green) expression of HUVECs at 3 days in microwell 8#, 18#, 36# and 46#. Cell nucleus, DAPI (blue). Scale bar = 200 μm. Microwell 1# and 8# represent G1 group; microwell 18# represent G2 group; microwell 28# represent G3 group; microwell 36# represent G4 group; and microwell 46# represent G5 group.

3.3 3D dynamic culture and phenotype of glioma cells

We also achieved 3D dynamic culture of glioblastoma U87-eGFP cells in homogeneous additive-free GelMA microgels, with results shown in Fig. 5-A and B. Since the cells inherently expressed green fluorescent protein, panoramic images displayed green fluorescence from the cells. Cell viability staining used only PI to stain dead cells, demonstrating over 95% survival rate after 3 days of dynamic culture. MTT assays also revealed that the A490 values at days 3 and 5 increased by 1.68-fold and 2.44-fold respectively compared to day 0, indicating good cell growth in 3D GelMA microgel dynamic culture. We extended the culture duration and performed immunofluorescence staining for VIM and α-SMA in U87-eGFP cells on day 7, as shown in Fig. 5-C. The results demonstrated that after 7 days of 3D dynamic culture, glioblastoma cells demonstrated strong VIM expression (red and merged yellow) at junctions adjacent to fluid perfusion channels, whereas α-SMA expression was observed only in a small subpopulation of cells (red and merged yellow). In the 3D dynamic culture system of glioblastoma, the significant expression of VIM suggests that tumor cells exhibit mesenchymal characteristics, potentially enhancing their invasive ability through EMT.³⁵ Fluid shear stress further induces cytoskeletal remodelling, leading to the upregulation of VIM as a key protein in mechanical stress adaptation.³⁶ Meanwhile, the sporadic expression of α-SMA indicates that some cells potentially acquired a pericyte-like phenotype, a feature closely associated with tumor angiogenesis. Mechanical forces in the dynamic culture may activate pathways such as YAP/TAZ or TGF-β, promoting the expression of contractility-related genes like α-SMA.³⁷ Particularly at the perfusion channel junctions, stronger mechanical stimulation may synergistically activate these pathways, collectively upregulating VIM and α-SMA, thereby enhancing the tumor cells' invasiveness and microenvironmental adaptability.

Fig. 5 Results of 3D dynamic culture of U87-eGFP cells in homogeneous additive-free GelMA microgels. (A) The panoramic fluorescence image of U87-eGFP-loaded GelMA microgels in a chip. Scale bar = 1 mm. (B) Results of cell viability and proliferation performed by PI staining and MTT assay. Green fluorescence represents U87-eGFP; red represents PI. Scale bar = 200 μm. (C) VIM and α-SMA expression of U87-eGFP cells at 7 days. Red represents VIM/α-SMA; yellow in merge panels indicates co-localization with GFP. Scale bar = 200 μm.

3.4 Self-assembly of endothelial–glioblastoma multicellular aggregates

Furthermore, to model cell–cell interactions within the TME, we established a 3D dynamic co-culture of glioblastoma U87-eGFP cells with vascular endothelial cells under FN concentration gradients. This system incorporated key TME elements, including multicellular components, biochemical cues (gradient FN), and biophysical cues (fluid shear stress), thereby constructing a multidimensional glioblastoma-on-a-chip platform. We performed panoramic fluorescence imaging of the chip, with results shown in Fig. 6-A. Endothelial cells were labelled with Cell Tracker™ Red CMFDA, while U87-eGFP cells inherently expressed green fluorescent protein (GFP). Through multiday observation, we discovered significant cellular self-assembly phenomena in the G1 and G2 groups after 3 and 5 days. Compared to the monoculture, endothelial cells did not spontaneously spread but instead formed distinct aggregates composed of both endothelial and glioblastoma cells aggregates at 5 days, as illustrated in Fig. 6-B and S3. We also quantified the percentage of occurrences, number and volume distribution of these multicellular aggregates in different groups. The volume of aggregates was approximated using the formula for a prolate spheroid, V = 1/2 × L × W², where L and W represent the major and minor axes, respectively. Results are shown in Fig. 6-C–F. The data demonstrated that the probability, quantity, and estimated volume of aggregates in the high FN concentration G1 were significantly higher than other groups, showing a gradual decrease with declining FN concentration in subsequent G2 and G3 groups. No multicellular aggregates were observed in G4 and G5 groups. Analysis of volume distribution for all detected aggregates showed relatively smooth probability before 1.4 × 10⁶ μm³, with an overall average estimated volume of 0.769 × 10⁶ μm³. These results indicated that the formation of endothelial–glioblastoma multicellular aggregates was closely associated with FN concentration in ECM. In the G1, G2, and G3 groups, vascular endothelial cells exhibit enhanced extension capacity and strengthened intercellular interactions, which might facilitate self-organization of multicellular aggregates. This phenomenon may originate from FN-mediated regulation of cytoskeletal reorganization and collective cell migration behavior through integrin-dependent cell adhesion. Simultaneously, FN may also alter the mechanical properties of the ECM, such as increasing stiffness, thereby synergistically promoting tumor–endothelial cell co-aggregation. These compact endothelial–glioblastoma aggregates observed in our model may represent an early, pre-angiogenic stage of tumor–vascular interaction, primarily driven by FN-mediated initial recognition, adhesion, and co-aggregation between the two cell types. This phase is critical in tumor progression, even prior to the formation of functional vasculature. Our findings align with previous reports on FN's role in promoting tumor cell aggregation and microenvironment remodelling. Specifically, we provide new evidence that FN concentration is a key regulator governing the 3D self-organization of endothelial–glioma cell aggregates. Although the current model does not recapitulate full angiogenesis or the formation of lumenized structures, the FN-dependent co-assembly demonstrated here may constitute a critical preliminary step in the complex process of tumor–vascular crosstalk.^38–40

Fig. 6 Results of self-assembly of endothelial–glioblastoma multicellular aggregates. (A) Schematic diagram and the panoramic fluorescence image of U87-eGFP cells cultured with HUVECs in a chip. Scale bar = 1 mm. (B) Fluorescence images of U87-eGFP cells cultured with HUVECs at 3 and 5 days in microwell 10# and 28#. HUVECs labelled with Cell Tracker™ Red CMFDA (red), and U87-eGFP cells expressed GFP (green). Scale bar = 200 μm. (C) Percentage of occurrences of multicellular aggregates at different groups. (D) Number of multicellular aggregates at different groups. (E) Volume of multicellular aggregates at different groups. (F) Volume of distribution of total multicellular aggregates.

To verify whether the self-organized endothelial–glioma multicellular aggregates possess cancer stem cell properties, we also examined relevant markers through immunofluorescence staining and real-time quantitative PCR. Compared with previous results (Fig. 5-C), when U87-eGFP cells alone underwent 3D dynamic culture for 7 days, there was minimal α-SMA expression. In contrast, for the co-culture of U87-eGFP cells and endothelial cells, significant α-SMA expression was observed as early as day 3 of co-culture, and it became more pronounced after the formation of multicellular aggregates (5 days), as shown in Fig. 7-A. In the fluorescence images, U87-eGFP cells appeared green due to their natural expression of green fluorescent protein, while α-SMA-positive staining was red. Endothelial cells remained unstained and were only visible in the bright field magnified insets. Additionally, we performed CD133 staining on the multicellular aggregates, with results displayed in Fig. 7-B. CD133 is a cell surface glycoprotein generally recognized as a characteristic marker of CSCs. In scientific research, CD133 expression levels are associated with tumor proliferation, metastasis, drug resistance, and other biological behaviours. We selected multicellular aggregates of varying sizes from the G1 group (microwell 8#) and G2 group (microwell 15#) for staining, both of which exhibited clear CD133 expression. This indicated the presence of CSCs in the self-organized endothelial–glioma multicellular aggregates, which exhibited key characteristics of tumor organoids. To further validate this conclusion, we prepared uniform GelMA microgels with a final FN concentration of 50 ng mL⁻¹ on the chip and compared the expression of VIM, α-SMA, and CD133 in total mRNA extracted from co-cultured cells on day 0 (control group) and day 5 (chip group). The experimental processes and results are shown in Fig. 7-C. The chip group demonstrated significant upregulation of VIM, α-SMA, and CD133 expression, with increases of 2.0-fold, 45.8-fold, and 559.0-fold compared to day 0 of co-culture on the chip, respectively. In the FN (50 ng mL⁻¹)-modified GelMA microgel system, the multicellular tumor aggregates formed after 5 days of dynamic culture may be enriched with cancer stem cell populations, as supported by the significant upregulation of key markers (VIM, α-SMA, CD133). The marked increase in CD133 suggests enrichment of CSCs, consistent with previous findings that CD133⁺ cells exhibit strong tumorigenicity and radio-resistance.⁴¹ The coordinated upregulation of VIM and α-SMA reflects EMT-stemness coupling and the acquisition of a pericyte-like phenotype.⁴² FN's pro-stemness effects are likely mediated through integrin signaling and ECM topological guidance.⁴³

Fig. 7 Expression and resistance analysis of endothelial–glioblastoma multicellular aggregates. (A) Fluorescence image of α-SMA expression of U87-eGFP cells co-cultured with HUVECs in a chip at 3 and 5 days. Green represents U87-eGFP; red indicates α-SMA; yellow in merged panels shows co-localization. Scale bar = 200 μm. Microwell 3# and 8# represent G1 group. The inset shows a magnified view of the bright field and fluorescence at the corresponding location. Scale bar = 50 μm. (B) Bright field and fluorescence images of CD133 expression of U87-eGFP cells co-cultured with HUVECs in a chip at 5 days. Microwell 8# represent G1 group; microwell 15# represent G2 group. Green represents U87-eGFP; blue indicates CD133. Scale bar = 50 μm. (C) Gene expression of cells. Control group, mixing cells were at 0 day; chip group, mixing cells were cultured in chips after 5 days. β-Actin refers to the internal parameter; *p < 0.05, **p < 0.01. (D) TMZ evaluation grouping and results of cytotoxity assay by MTT. ***p < 0.001.

The formation of endothelial–glioma multicellular aggregates and the upregulation of markers such as CD133 are hallmarks of stemness and are closely correlated with drug resistance. Based on this, we evaluated the response to the antitumor drug temozolomide.^44,45 Temozolomide is an oral chemotherapeutic agent commonly used for glioblastoma treatment, which can spontaneously hydrolysis to produce the active metabolite MTIC and exert antitumor effects. Literature reports indicated that the IC₅₀ values of temozolomide for U87-eGFP cells mostly range between 30–300 μM. Using the traditional 96-well plate-based MTT assay, we determined the IC₅₀ of temozolomide for U87-eGFP cells in our experiments (drug exposure time: 2 days) to be approximately 154 μM, as shown in Fig. S4. Based on this, we performed drug evaluation under 3D dynamic culture conditions on a chip, divided into four groups: U87-eGFP cells in uniform GelMA microgels without drug treatment (U87-eGFP-no drug group); U87-eGFP cells in uniform GelMA microgels treated with 150 μM temozolomide for two days (U87-eGFP-TMZ group); U87-eGFP cells co-cultured with HUVECs in uniform GelMA microgels containing 50 ng mL⁻¹ FN without drug treatment (U87-eGFP + HUVECs-no drug group); and U87-eGFP cells co-cultured with HUVECs in uniform GelMA microgels containing 50 ng mL⁻¹ FN treated with 150 μM temozolomide for two days (U87-eGFP + HUVECs-TMZ group). The grouping and results are shown in Fig. 7-D. The results indicated that the cell viability of U87-eGFP cells in the U87-eGFP-TMZ group was 59% compared to the U87-eGFP-no drug group, suggesting that the 3D dynamic environment may induce drug resistance in U87-eGFP cells. On the other hand, the overall cell viability in the U87-eGFP + HUVECs-TMZ group was 72% of that in the U87-eGFP + HUVECs-no drug group. This further increase suggested enhanced resistance to temozolomide, which may be closely related to elevated CD133 expression levels and the formation of tumor organoid-like aggregates. The 3D culture environment enhances tumor cell drug resistance through ECM-mediated mechanisms. More importantly, endothelial cell co-culture promotes the formation of specialized glioblastoma aggregates that not only enrich CSCs (as evidenced by upregulated CD133 expression) but also maintain their stem-like properties through secretion of various cytokines, thereby further enhancing overall drug resistance. These findings confirm the utility of the 3D dynamic culture system and, more importantly, highlight the crucial role of tumor–endothelial cell interactions in the development of drug resistance.

4. Conclusions

We developed a glioblastoma-on-a-chip platform featuring FN gradient-GelMA microgel to investigate self-organized endothelial–glioma multicellular aggregates and tumor stem cell differentiation. The established glioblastoma-on-a-chip integrates key TME cues, including cell–cell interactions, ECM biochemical gradients (e.g., FN), and dynamic biophysical forces (e.g., fluid shear stress). We demonstrated that both glioblastoma and endothelial cells maintained excellent viability and proliferative capacity in this system. Crucially, the FN concentration effectively regulated the formation of endothelial–glioblastoma multicellular aggregates. At the same time as the appearance of multicellular aggregates, there was a significant increase in the expression of tumor related markers such as α-SMA, VIM, and CD133, which indicated the emergence of tumor stem cells and might lead to tumor drug resistance. We compared its toxicity results with traditional methods for anti-glioma chemotherapy drug temozolomide and confirmed the emergence of drug resistance. This established tumor-on-a-chip proposes an integrated concentration gradient generation and solidification strategy based on the microfluidic chip structure without complex external driving. It successfully integrates key characteristics that are difficult to achieve simultaneously in traditional technologies, such as wide-range continuity, precise quantification ability, excellent biocompatibility, and high-throughput parallelization, into a simple and integrated platform, providing a promising technical tool for in vitro biological model construction and efficient drug screening.

Author contributions

Jianing Li: investigation, conceptualization, and writing – original draft. Xinghua Gao: funding acquisition, project administration, supervision, writing – original draft and writing – review and editing. Xiaoling Yang: investigation, conceptualization and visualization. Hongcai Wang: investigation, and conceptualization. Xindi Sun: investigation, and visualization. Chang Xue: investigation and validation. Jingyun Ma: funding acquisition, project administration, supervision, and writing – review and editing.

Conflicts of interest

The authors declare no competing interests.

Data availability

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

Supplementary information (SI): Fig. S1 FTIR spectroscopy of GelMA microgel (red) and pure GelMA (blue). Fig. S2. Standard curve of concentration versus relative fluorescence intensity for GelMA microgels loaded with different concentrations of FITC-dextran. Fig. S3. Fluorescence and bright-field images of U87-eGFP cells cultured with HUVECs at 3 and 5 days in microwell 7#, 10#, 18# and 28#. Fig. S4. Cytotoxicity assay of U87-eGFP cells treated with temozolomide (TMZ) for 48 h. See DOI: https://doi.org/10.1039/d5lc01018g.

Acknowledgements

This work was supported by Key Program of the National Natural Science Foundation of China (No. 12332016), Zhejiang Provincial Natural Science Foundation of China (grant no. LTGY24B050001), Zhejiang Medicine and Health Science and Technology Program (grant no. 2024KY283), Ningbo Natural Science Foundation (grant no. 2022J252), Shanghai Science and Technology Commission (24ZR1460300).

References

1. J. White, M. P. J. White, A. Wickremesekera, L. F. Peng and C. Gray, The tumour microenvironment, treatment resistance and recurrence in glioblastoma, J. Transl. Med., 2024, 22, 540.
2. S. Singh, D. Dey, D. Barik, I. Mohapatra, S. Kim, M. Sharma, S. Prasad, P. Z. Wang, A. Singh and G. Singh, Glioblastoma at the crossroads: current understanding and future therapeutic horizons, Signal Transduction Targeted Ther., 2025, 10, 213.
3. A. Bikfalvi, C. A. da Costa, T. Avril, J. V. Barnier, L. Bauchet, L. Brisson, P. F. Cartron, H. Castel, E. Chevet, H. Chneiweiss, A. Clavreul, B. Constantin, V. Coronas, T. Daubon, M. Dontenwill, F. Ducray, N. Entz-Werlé, D. Figarella-Branger, I. Fournier, J. S. Frenel, M. Gabut, T. Galli, J. Gavard, G. Huberfeld, J. P. Hugnot, A. Idbaih, M. P. Junier, T. Mathivet, P. Menei, D. Meyronet, C. Mirjolet, F. Morin, J. Mosser, E. C. J. Moyal, V. Rousseau, M. Salzet, M. Sanson, G. Seano, E. Tabouret, A. Tchoghandjian, L. Turchi, F. M. Vallette, S. Vats, M. Verreault and T. Virolle, Challenges in glioblastoma research: focus on the tumor microenvironment, Trends Cancer, 2023, 9, 9–27.
4. S. Maity, C. Jewell, C. Yilgor, S. Kawakita, S. Sharma, A. Gomez, M. Mecwan, N. Falcone, M. Ermis, M. Monirizad, N. H. Kouchehbaghi, F. Zehtabi, D. Khorsandi, M. R. Dokmeci, D. Moniz-Garcia, A. Quiñones-Hinojosa, A. Khademhosseini and V. Jucaud, Deciphering pericyte-induced temozolomide resistance in glioblastoma with a 3D microphysiological system mimicking the biomechanical properties of brain tissue, Acta Biomater., 2025, 200, 202–217.
5. N. M. Anderson and M. C. Simon, The tumor microenvironment, Curr. Biol., 2020, 30, R921–R925.
6. Z. W. Hu, Y. X. Cao, E. A. Galan, L. Hao, H. R. Zhao, J. Y. Tang, G. Sang, H. Q. Wang, B. Xu and S. H. Ma, Vascularized Tumor Spheroid-on-a-Chip Model Verifies Synergistic Vasoprotective and Chemotherapeutic Effects, ACS Biomater. Sci. Eng., 2022, 8, 1215–1225.
7. T. Wenta, P. Nastaly, B. Lipinska and A. Manninen, Remodeling of the extracellular matrix by serine proteases as a prerequisite for cancer initiation and progression, Matrix Biol., 2024, 134, 197–219.
8. D. Pally and A. Naba, Extracellular matrix dynamics: A key regulator of cell migration across length-scales and systems, Curr. Opin. Cell Biol., 2024, 86, 102309.
9. X. J. Chu, W. T. Tian, J. Y. Ning, G. Xiao, Y. Q. Zhou, Z. Q. Wang, Z. F. Zhai, G. Tanzhu, J. Yang and R. R. Zhou, Cancer stem cells: advances in knowledge and implications for cancer therapy, Signal Transduction Targeted Ther., 2024, 9, 170.
10. X. Y. Ding, Y. Y. Liang, S. Y. Zhou, Y. Wu, P. Salata, A. Mikolajczk-Martinez, V. Khosrawipour and Z. W. Zhang, Targeting tumor extracellular matrix with nanoparticles to circumvent therapeutic resistance, J. Controlled Release, 2025, 383, 113786.
11. J. J. F. Sleeboom, G. S. van Tienderen, K. Schenke-Layland, L. J. W. van der Laan, A. A. Khalil and M. M. A. Verstegen, The extracellular matrix as hallmark of cancer and metastasis: From biomechanics to therapeutic targets, Sci. Transl. Med., 2024, 16, eadg3840.
12. J. Y. Ma, N. Li, Y. C. Wang, L. Wang, W. J. Wei, L. M. Shen, Y. Sun, Y. Jiao, W. G. Chen and J. Liu, Engineered 3D tumour model for study of glioblastoma aggressiveness and drug evaluation on a detachably assembled microfluidic device, Biomed. Microdevices, 2018, 20, 80.
13. D. D. Shah, N. R. Raghani, M. R. Chorawala, S. Singh and B. G. Prajapati, Harnessing three-dimensional (3D) cell culture models for pulmonary infections: State of the art and future directions, Naunyn-Schmiedeberg's Arch. Pharmacol., 2023, 396, 2861–2880.
14. R. J. Xie, V. Pal, Y. R. Yu, X. L. Lu, M. W. Gao, S. J. Liang, M. Huang, W. J. Peng and I. T. Ozbolat, A comprehensive review on 3D tissue models: Biofabrication technologies and preclinical applications, Biomaterials, 2024, 304, 122408.
15. K. Kwapiszewska, A. Michalczuk, M. Rybka, R. Kwapiszewski and Z. Brzózka, A microfluidic-based platform for tumour spheroid culture, monitoring and drug screening, Lab Chip, 2014, 14, 2096–2104.
16. H. T. Liu, Y. Q. Wang, K. L. Cui, Y. Q. Guo, X. Zhang and J. H. Qin, Advances in Hydrogels in Organoids and Organs-on-a-Chip, Adv. Mater., 2019, 31, 1902042.
17. R. Gauvin, Y. C. Chen, J. W. Lee, P. Soman, P. Zorlutuna, J. W. Nichol, H. Bae, S. C. Chen and A. Khademhosseini, Microfabrication of complex porous tissue engineering scaffolds using 3D projection stereolithography, Biomaterials, 2012, 33, 3824–3834.
18. C. He, P. Feng, D. Xu, L. Lian, J. Li, K. Gan, Y. Gu, T. Pan, Y. Tang, X. Wu, Z. Peng, C. Ke and J. Ma, Naringenin-Loaded Core–Shell Mg-MOF Nanocomposite Integrated Injectable Hydrogel for Bone Regeneration: Anti-inflammatory, Osteogenic, and Angiogenic, Biomacromolecules, 2025, 26, 7706–7723.
19. P. P. Feng, C. N. He, G. R. Li, J. Li, Y. Luo, Y. Q. Chen, Y. Tang, J. Y. Ma and C. H. Ke, A light-cured injectable composite hydrogel based on chitosan and decellularized matrix modulates stem cell aggregation behavior for accelerating cartilage defect repair, Int. J. Biol. Macromol., 2025, 295, 139711.
20. J. Bae, H. Jeon and T. Kim, Full-Combinatorial Concentration Gradient Array with 3D Micro/Nanofluidics for Antibiotic Susceptibility Testing, Anal. Chem., 2024, 96, 5462–5470.
21. H. Y. Lee, Y. Y. Choi, J. W. Kim, J. W. Choi, Y. K. Kim and B. G. Chung, Enhancement of Maturation and Functionality of Human Fabry Disease Kidney Organoids in a Microfluidic Gradient Chip, BioChip J., 2025, 19, 152–165.
22. S. Maity, T. Bhuyan, C. Jewell, S. Kawakita, S. Sharma, H. T. Nguyen, A. H. Najafabadi, M. Ermis, N. Falcone, J. J. Chen, K. Mandal, D. Khorsandi, C. Yilgor, A. Choroomi, E. Torres, M. Mecwan, J. V. John, M. Akbari, Z. H. Wang, D. Moniz-Garcia, A. Quiñones-Hinojosa, V. Jucaud, M. R. Dokmeci and A. Khademhosseini, Recent Developments in Glioblastoma-On-A-Chip for Advanced Drug Screening Applications, Small, 2025, 21 DOI:10.1002/smll.202405511.
23. M. Ustun, S. R. Dabbagh, I. S. Ilci, T. Bagci-Onder and S. Tasoglu, Glioma-on-a-Chip Models, Micromachines, 2021, 12, 490.
24. X. L. Yang, J. Y. Shi, B. R. Shi, J. N. Li, C. Xue, J. Y. Ma and X. H. Gao, Micro- and nano-fibers for organ-on-a-chip: Construction, applications, and prospects, Mater. Today Bio, 2024, 29, 101322.
25. X. Cui, C. Ma, V. Vasudevaraja, J. Serrano, J. Tong, Y. S. Peng, M. Delorenzo, G. M. Shen, J. Frenster, R. T. T. Morales, W. Y. Qian, A. Tsirigos, A. S. Chi, R. Jain, S. C. Kurz, E. P. Sulman, D. G. Placantonakis, M. Snuderl and W. Q. Chen, Dissecting the immunosuppressive tumor microenvironments in Glioblastoma-on-a-Chip for optimized PD-1 immunotherapy, eLife, 2020, 9, e52253.
26. F. F. Vit, Y. T. Wu, E. Fujiwara, H. F. Carvalho and L. G. D. la Torre, Microfluidic chip for synergic drugs assay in 3D breast cancer cell, Microfluid. Nanofluid., 2024, 28, 26.
27. J. Y. Shi, J. N. Li, W. T. Su, C. Xue, Y. Zhang and X. H. Gao, Engineered microenvironments and pancreatic islet-on-chips for screening sugar substitute and antidiabetic compounds, Food Res. Int., 2024, 196, 115084.
28. A. H. Alves, M. P. Nucci, J. B. Mamani, N. M. E. Valle, E. F. Ribeiro, G. N. A. Rego, F. A. Oliveira, M. H. Theinel, R. S. Santos and L. F. Gamarra, The Advances in Glioblastoma On-a-Chip for Therapy Approaches, Cancers, 2022, 14, 869.
29. J. Shi, J. Li, J. Ma, X. Yang, C. Xue, Y. Zhang and X. Gao, Engineering pancreatic islet-loaded microfibers via pneumatically-controlled microfluidic spinning for the assembly of a microphysiological system, Biomaterials, 2026, 324, 123480.
30. K. Yao, W. Li, K. Y. Li, Q. R. Wu, Y. R. Gu, L. J. Zhao, Y. Zhang and X. H. Gao, Simple Fabrication of Multicomponent Heterogeneous Fibers for Cell Co-Culture via Microfluidic Spinning, Macromol. Biosci., 2020, 20, 1900395.
31. J. F. Tan, L. Q. Zhu, J. Y. Shi, J. H. Zhang, J. Kuang, Q. W. Guo, X. J. Zhu, Y. L. Chen, C. B. Zhou and X. H. Gao, Evaluation of drug resistance for EGFR-TKIs in lung cancer via multicellular lung-on-a-chip, Eur. J. Pharm. Sci., 2024, 199, 106805.
32. L. L. Tian, J. Y. Shi, W. Li, Y. Zhang and X. H. Gao, Hollow Microfiber Assembly-Based Endocrine Pancreas-on-a-Chip for Sugar Substitute Evaluation, Adv. Healthcare Mater., 2024, 13 DOI:10.1002/adhm.202302104.
33. X. Y. Du, Q. Li, G. Wu and S. Chen, Multifunctional Micro/Nanoscale Fibers Based on Microfluidic Spinning Technology, Adv. Mater., 2019, 31, 1903733.
34. X. Liu, X. H. Wang, L. M. Zhang, L. L. Sun, H. R. Wang, H. Zhao, Z. T. Zhang, W. L. Liu, Y. M. Huang, S. Ji, J. J. Q. Zhang, K. Li, B. B. Song, C. Li, H. Zhang, S. Li, S. Wang, X. F. Zheng and Q. Gu, 3D Liver Tissue Model with Branched Vascular Networks by Multimaterial Bioprinting, Adv. Healthcare Mater., 2021, 10, 2101405.
35. D. H. L. Leung, B. W. S. Phon, M. Sivalingam, A. K. Radhakrishnan and M. N. A. Kamarudin, Regulation of EMT Markers, Extracellular Matrix, and Associated Signalling Pathways by Long Non-Coding RNAs in Glioblastoma Mesenchymal Transition: A Scoping Review, Biology, 2023, 12, 818.
36. N. C. A. van Engeland, F. S. Rodriguez, A. Rivero-Müller, T. Ristori, C. L. Duran, O. Stassen, D. Antfolk, R. C. H. Driessen, S. Ruohonen, S. T. Ruohonen, S. Nuutinen, E. Savontaus, S. Loerakker, K. J. Bayless, M. Sjöqvist, C. V. C. Bouten, J. E. Eriksson and C. M. Sahlgren, Vimentin regulates Notch signaling strength and arterial remodeling in response to hemodynamic stress, Sci. Rep., 2019, 9, 12415.
37. F. S. Gallardo, M. Cruz-Soca, A. Bock-Pereda, J. Faundez-Contreras, C. Gutiérrez-Rojas, A. Gandin, V. Torresan, J. C. Casar, A. Ravasio and E. Brandan, Role of TGF-β/SMAD/YAP/TAZ signaling in skeletal muscle fibrosis, Am. J. Physiol., 2025, 328, C1015–C1028.
38. J. W. Rick, A. Chandra, C. D. Ore, A. T. Nguyen, G. Yagnik and M. K. Aghi, Fibronectin in malignancy: Cancer-specific alterations, protumoral effects, and therapeutic implications, Semin. Oncol., 2019, 46, 284–290.
39. G. Efthymiou, A. Saint, M. Ruff, Z. Rekad, D. Ciais and E. Van Obberghen-Schilling, Shaping Up the Tumor Microenvironment With Cellular Fibronectin, Front. Oncol., 2020, 10, 641.
40. J. L. Wu, C. F. Xu, X. H. Yang and M. S. Wang, Fibronectin promotes tumor progression through integrin αvβ3/PI3K/AKT/SOX2 signaling in non-small cell lung cancer, Heliyon, 2023, 9, e20185.
41. P. M. Glumac and A. M. LeBeau, The role of CD133 in cancer: a concise review, Clin. Transl. Med., 2018, 7, 18.
42. S. Usman, N. H. Waseem, T. K. N. Nguyen, S. Mohsin, A. Jamal, M. T. Teh and A. Waseem, Vimentin Is at the Heart of Epithelial Mesenchymal Transition (EMT) Mediated Metastasis, Cancers, 2021, 13, 4985.
43. C. S. Wu, S. M. Weis and D. A. Cheresh, Upregulation of fibronectin and its integrin receptors - an adaptation to isolation stress that facilitates tumor initiation, J. Cell Sci., 2023, 136, jcs261483.
44. M. Amereh, S. Shojaei, A. Seyfoori, T. Walsh, P. Dogra, V. Cristini, B. Nadler and M. Akbari, Insights from a multiscale framework on metabolic rate variation driving glioblastoma multiforme growth and invasion, Commun. Eng., 2024, 3, 176.
45. M. Amereh, A. Seyfoori, S. Shojaei, S. Lane, T. Zhao, M. S. Barough, J. J. Lum, P. Walter and M. Akbari, Tumoroid Model Reveals Synergistic Impairment of Metabolism by Iron Chelators and Temozolomide in Chemo-Resistant Patient-derived Glioblastoma Cells, Adv. Sci., 2025, 12 DOI:10.1002/advs.202412505.

---

Footnote

† Co-first author: Jianing Li and Xinghua Gao have equal contribution for this work.

---