A vitronectin-functionalized ECM-mimetic hydrogel platform for modelling integrin-dependent tumour responses to cilengitide Authors: Isaac Vieco-Martí, Sofia Granados-Aparici, Amparo López-Carrasco, Miguel F. Segura, Manuel Salmerón-Sánchez, Inmaculada Noguera, Samuel Navarro, Rosa Noguera Published: 2026 ABSTRACT The extracellular matrix (ECM) critically regulates tumour cell adhesion, mechanotransduction and therapeutic response, yet there is a lack of in vitro models that faithfully reproduce ECM–integrin signalling while remaining compatible with histological and molecular workflows. Here, we present a biomimetic hydrogel platform based on gelatin–tyramine and silk fibroin, engineered to recreate defined vitronectin microenvironments and integrin-targeting conditions, including pharmacological inhibition with the αvβ3-integrin antagonist cilengitide. The platform enables control over matrix composition and bioactivity while preserving tissue-like architecture. We establish robust protocols for hydrogel synthesis under distinct vitronectin conditions, cell embedding, pharmacological treatment and downstream processing, including genomic DNA isolation. In addition, the platform supports standard histological processing and immunostaining, enabling digital pathology-based assessment of morphological features, vitronectin expression patterns, and adhesion-related markers such as vinculin and β3 integrin. Importantly, the system enables recovery of viable cells, allowing subsequent orthotopic implantation in vivo. Together, these workflows provide a reproducible and scalable framework to investigate ECM-mediated regulation of neuroblastoma cell behaviour and drug response, with scope for quantitative image analysis of integrin-associated signals and ECM deposition. This ECM-mimetic hydrogel system offers a versatile and histology-compatible experimental platform for studying integrin-dependent tumour–matrix interactions and developing physiologically relevant in vitro–in vivo translational pipelines. Introduction The extracellular matrix (ECM) is a dynamic compartment of the tumour microenvironment composed of adhesion glycoproteins and fibrous proteins that provides structural support and mechanical cues [[1], [2], [3], [4]]. Alterations in ECM architecture and stiffness, generally measured by Young's modulus, can promote aggressive tumour phenotypes through mechanotransduction at the cell–ECM interface [[5], [6], [7], [8]]. A major component of this signalling involves integrins, including αvβ3, which binds RGD motifs present in multiple ECM proteins and has been implicated in tumour progression, angiogenesis, invasion, and therapy resistance, making it a relevant mechanobiology-linked therapeutic target [5,6]. These processes involve the integration of mechanical signals into transcriptional responses through mechanosensitive pathways such as YAP/TAZ, which regulate proliferation, survival and differentiation in response to matrix stiffness and adhesion cues. In several neoplasms, ECM stiffness has been directly linked to YAP-driven transcriptional programmes associated with tumour progression [9]. However, mechanotherapy strategies that target mechanotransduction nodes such as integrins face important translational limitations, including ligand and receptor redundancy, and compensatory adhesion pathways that can reduce efficacy [[10], [11], [12], [13]]. Clinical implementation of these drugs has also been hindered by the lack of robust biomarkers to identify responsive contexts and unexpected activation of mechanisms [[14], [15], [16]]. Tuneable 3D models can mimic tissue-like organisation and enable controlled matrix composition, thereby supporting the study of cell–ECM-dependent drug responses and helping identify potential biomarkers of activity. In contrast, conventional 2D culture systems fail to recapitulate these biomechanical and spatial constraints, often leading to altered cell morphology, polarity and signalling, including dysregulated integrin engagement and mechanotransduction pathways [17]. This limitation is particularly relevant when studying integrin-targeted therapies, where ligand availability and spatial organisation critically determine functional outcomes [5]. Among the available 3D platforms, gelatin–tyramine (GTA) and silk fibroin (sf) offer a practical balance of biological relevance and experimental simplicity. GTA adds cell-interactive cues and sf is a biocompatible, biodegradable biomaterial widely used for 3D culture [[18], [19], [20]]. Enzymatic dityrosine crosslinking using horseradish peroxidase and low concentrations of H2O2 enables gelation at 37 °C under mild conditions that do not compromise cell viability, with Young's modulus in the range relevant, among other tissues, for nervous tissue [[20], [21], [22], [23]]. The same chemistry also allows straightforward incorporation of additional tyrosine-containing ECM glycoproteins without modification steps [20]. Our focus is to study GTA-sf HGs in the specific setting of neuroblastoma, the most common extracranial solid tumour of childhood which accounts for 15% of childhood cancer-related deaths [24]. Despite major therapeutic advances, high-risk neuroblastoma (HR-NB) disease still carries poor long-term survival and frequent relapse, highlighting the need for new biomarkers and more effective targeted strategies, including approaches that consider the tumour microenvironment [25]. In this context, vitronectin (VN) is an RGD-containing ECM glycoprotein that binds integrins such as αvβ3 and αvβ5 and has been implicated in the aggressiveness of multiple and different tumoral pathologies [[26], [27], [28]]. Our group has shown that VN is expressed in HR-NB patient samples, murine models and multiple 3D models. VN is associated with aggressive clinicopathological and genomic features by digital pathology and is detectable as a secreted factor in culture media and patient plasma, supporting its relevance as a biomarker and a candidate therapeutic target in HR-NB [[29], [30], [31], [32]]. In this system, therefore, GTA-sf provides a permissive structural matrix, while VN serves as the main instructive cue examined in relation to integrin-dependent behaviour. Using the GTA–sf platform with variable VN microenvironments, we have already demonstrated HR-NB compatibility and applied short-term readouts including real-time holographic microscopy, as well as ELISA-based detection of VN secretion into culture media [32,33]. Here, we show new applications of this work by detailing GTA–sf HG fabrication and presenting a set of workflows for multi-level characterisation, including mechanical testing, formalin-fixed paraffin embedded compatible histopathology and imaging, genomic profiling, and downstream cell recovery for in vivo implantation, while also illustrating its qualitative potential for drug testing in defined VN microenvironments. Cilengitide (CLG) is a cyclic RGD peptide targeting αvβ3 and αvβ5 integrins which has been clinically evaluated in a phase III clinical trial for glioblastoma but has shown limited efficacy and a lack of response biomarkers [7,34,35]. Our prior data nonetheless suggest that CLG remains of interest in HR-NB, showing cytotoxicity in vitro, synergy with etoposide (chemotherapy used in rapid COJEC regime treatment for HR-NB patients [36]), signals of reduced αvβ3 expression by immunohistochemistry assay and growth restraint in a preliminary in vivo setting [31,37]. These observations form the rationale for proof-of-concept testing of CLG in a 3D context with controllable VN availability. In this context, the non-VN-added condition models exposure to endogenous VN, whereas the VN-rich condition represents increased extracellular VN availability. The platform does not aim to replicate full tissue complexity, but rather to provide mechanistic resolution of specific tumour–matrix interactions within a controllable experimental framework. This is biologically relevant because VN-rich patterns in neuroblastoma are linked to aggressive disease, and host VN status can influence tumour selection in orthotopic models [30,38]. Thus, the VN-functionalized GTA–sf HG platform provides a 3D system for studying integrin-dependent morphology-based tumour phenotypes and their modulation by CLG using HR-NB as a representative tumour example. Materials and methods 2D culture and PDX suspension culture of commercial cell lines We used MYCN-amplified SK-N-BE(2) and ALK-mutated SH-SY5Y human cell lines as models of HR-NB [39,40]. Both SK-N-BE(2) and SH-SY5Y were acquired from American Type Culture Collection (ATCC, Manassas, VA, USA), expanded on Iscove's Modified Dulbecco's Medium (IMDM, Gibco, Thermofisher), supplemented with 10% foetal bovine serum (FBS, Thermofisher), 1% Insulin-Transferrin-Selenium G Supplement (Thermofisher), 0.2% Plasmocin (InvivoGen) and 1% penicillin/streptomycin (Thermofisher). Cells were maintained at 37 °C and 5% CO2 atmosphere. Media was replaced every two days. In addition to commercial cell lines, we included two previously established and published patient-derived xenograft (PDX) models from patients at Hospital Clínico Universitario de València (PDX1 and PDX2) [41,42]. Both PDX cell lines (MYCN-amplified) were expanded in suspension on DEMEM-Glutamax without pyruvate (Gibco, Thermofisher), F12 Nutrimix Glutamax (Gibco, Thermofisher), 2% B27 without vitamin A (Thermofisher), 1% penicillin/streptomycin (Thermofisher), 40 ng/ml bFGF (Peprotech) and 20 ng/ml EGF (Peprotech). Cells were maintained at 37 °C and 5% CO2 atmosphere. Media was replaced every two days. 3D hydrogel fabrication protocol and CLG treatment Dityrosine bond formation is a key mechanism used for HG polymerisation at 37 °C. Phenol groups were introduced into gelatin to enhance its ability to form dityrosine bonds, resulting in gelatin–tyramine (GTA), as previously described [43]. Briefly, gelatin (Sigma Aldrich, St. Louis, MO) was melted in 50 mL of 0.05 M 2-(N-morpholino) ethane sulfonic acid buffer at 60 °C and then cooled to 25 °C. Subsequently, 500 mg of tyramine (Sigma Aldrich, St. Louis, MO) was added together with 184 mg of carbodiimide hydrochloride and 57 mg of N-hydroxysuccinimide to facilitate the carbodiimide-mediated reaction. The mixture was agitated for 18 h at room temperature, followed by dialysis against distilled water using a 3.5 molecular weight cut-off membrane and lyophilisation. The GTA was stored at −20 °C until use. HGs were prepared with a final volume of 65 μL. SK-N-BE(2) and SH-SY5Y cells were embedded at a target density of 2 × 106 viable cells mL (approximately 120,000 cells per HG), while PDX1 and PDX2 cells were embedded at 4 × 106 viable cells mL (approximately 240,000 cells per HG). The HG fabrication stock solutions, HG volumes, and final component concentrations are summarised in Table 1.Table 1Mix of compounds for GTA-sf HGs.Table 1Mix of compounds for GTA-sf HGsCompound StockμL/HGFinal ConcentrationGTA (50 mg/mL)12.5 μL10 mg/mLsf (50 mg/mL)37.5 μL30 mg/mlVN (2 mg/mL) or dPBS12.5 μL0.4 mg/mLHRP (1000 U/mL)1.25 μL20 U/mLH2O2 (0.3%)2 μL0.01 %Stock concentrations, amount of each stock needed per HG of 65 μL and resulting final concentrations after mixing. Table 1Mix of compounds for GTA-sf HGs.Table 1Mix of compounds for GTA-sf HGsCompound StockμL/HGFinal ConcentrationGTA (50 mg/mL)12.5 μL10 mg/mLsf (50 mg/mL)37.5 μL30 mg/mlVN (2 mg/mL) or dPBS12.5 μL0.4 mg/mLHRP (1000 U/mL)1.25 μL20 U/mLH2O2 (0.3%)2 μL0.01 %Stock concentrations, amount of each stock needed per HG of 65 μL and resulting final concentrations after mixing. Mix of compounds for GTA-sf HGs. Fabrication was performed following a previously reported protocol [20]. Lyophilized GTA was weighed, UV sterilised for 10 min and melted in serum-free IMDM in a 37 °C water bath to obtain a stock solution of 50 mg/mL. GTA was mixed with sf solution (50 mg mL, Sigma Aldrich, Merck, USA) at a ratio of 25:75. Following trypsinization, viable SK-N-BE(2) and SH-SY5Y cells were counted with trypan blue assay and resuspended in the GTA–sf mixture to achieve the target density. For PDX1 and PDX2, spheroids were dissociated using Accutase (Sigma Aldrich, Merck, USA), and after 12 h in culture medium, viable cells were counted with trypan blue assay and resuspended in the GTA–sf mixture to achieve the target density. For VN-rich HGs, VN (PrepoTech, London, UK) was resuspended in dPBS to a stock concentration of 2 mg/mL. The required volume to achieve the final concentration indicated in Table 1 was placed in a tube; for non-VN-added HGs we used an equivalent volume of dPBS. The corresponding volume of horseradish peroxidase (HRP) stock solution (type VI, Sigma Aldrich, St. Louis, MO) was then added to the VN or dPBS tube to obtain a final concentration of 20 U mL (double that of the original protocol to accelerate enzymatic crosslinking kinetics without altering structural properties [20,44,45]). The GTA–sf–cell suspension was subsequently added to the VN/dPBS–HRP tube and gently mixed. All components except hydrogen peroxide were therefore combined prior to polymerisation. Polymerisation was initiated by placing 2 μL of hydrogen peroxide stock solution (0.3%) in the centre of a 24-well plate, followed by gentle addition of 63 μL of the GTA–sf–VN/dPBS–cell to avoid premature polymerisation during handling. HGs were incubated for 45 min at 37 °C to allow complete enzymatic polymerisation and subsequently overlaid with 2 mL of the corresponding culture medium. Cells were cultured at 37 °C in 5% CO2 for two or three weeks, with medium changes every three days. Cilengitide trifluoroacetic acid (CLG) [salt ≥98% (HPLC), SML1594-25 MG, Merck] was dissolved in dPBS. A subset of HGs was treated with six CLG doses of 100 μM. Addition of CLG started after the first change of media, and subsequent dosages were applied at every change of media for up to three weeks. Measurement of hydrogels elastic moduli Elastic moduli of non-treated HGs for all cell lines were measured using a mechanical compression test performed with a Zwick Roell Z0.5 system. Following two or three weeks of culture, HGs were placed on a microscope slide, and their width, height and depth were measured with a ruler. A subset of HGs of both VN conditions was crafted without any cells to use as reference for mechanical compression tests. HG morphology was approximated as a cube. The elastomer was set to a pre-charge of 3 mN and compressive force was applied to a speed of 20% of deformation/min, until reaching 30% of deformation. The device recorded the force deformation curve. Measurements showing non-continuous curves were discarded. When possible, three measurements per HG were performed. Elastic moduli, expressed in kPa, were calculated using testXpert III software (Zwick Roell) by linear regression between 5% and 15% deformation, corresponding to the linear region of the force-deformation curve. Data was exported from testXpert III (Zwick Roell software) to R. The curve plots were grouped by time of culture (two and three weeks) and VN addition to the matrix (VN-rich and non-VN-added). To test differences between VN-rich vs. non-VN-added conditions, the two-tailed non-parametric Wilcoxon rank-sum test was used. A p-value threshold of <0.05 was considered statistically significant. Comparisons were predefined based on biological rationale; no multiple testing correction was applied. Hydrogel histological processing and staining At the 2- and 3-week time points HGs were formalin-fixed, paraffin-embedded (FFPE), then cut into 3 μm sections. HG sections were stained with Haematoxylin & Eosin (H&E) and immunoassayed against VN (EP873Y, Clone; ab45139, Abcam, Cambridge, MA, USA) diluted at 1:100, using OptiView Amplification Kit (Ventana Medical Systems Inc., Tucson, U.S.A.) in the BenchMark XT automated slide staining system (Ventana Medical Systems Inc., Tucson, USA). Two tissue microarrays (TMA) were built for super-resolution microscopy. TMA cores of 1.2 mm were extracted from FFPE blocks including tissue from SK-N-BE(2), SH-SY5Y, PDX1 and PDX2 for both VN conditions. TMA was cut in into 3 μm sections for fluorescence staining. Samples were deparaffined using a battery of decreasing alcohol concentrations, and antigen retrieval was performed by heating samples for 20 min in the microwave in citrate buffer 10 mM PH 6.0. The slides were then rinsed in water and incubated in 20 mM glycine solution for 5 min to reduce quenching. After three 5-min washes in PBS, the slides were incubated three times for 5 min with 0.3% PBS-TritonX100 and then blocked with 1% BSA solution for 1 h. The samples were incubated overnight at 4 °C with antibodies against vinculin (V9131 Mouse, Merck) and integrin β3 (D7X3P Rabbit, Cell Signaling) diluted to 1:200 in 1% BSA solution. After three washes with PBST 0.5%, the samples were incubated with secondary antibodies with TexasRed at 1:100 dilution (Horse anti-mouse IgG Antibody [H + L], TI-2000 Vector Laboratories) and Alexa488 at 1:250 dilution (Donkey anti-rabbit IgG Antibody [H + L], Invitrogen) for 1 h. After three 5-min washes with PBST 0.5%, the samples were incubated with DAPI (Invitrogen™ D1306,1 μg/mL) for 5 min, rinsed with PBS and mounted with antifading media (Fluoromount-G™ Mounting Medium). Slide digitisation, hydrogel area determination and fluorescence image post-processing H&E and anti-VN slides were scanned with Ventana iScanHT (Roche) at 20x using the HT strategy, resulting in TIF format RGB images at 0.465 μm/pixel resolution. A subset of slides was scanned with 3D Histech PANNORAMIC 250 Flash III 40x, obtaining TIF format RGB images at a resolution of 0.25 μm/pixel. Digital slides were reviewed in QuPath (version 0.6) [46] by an experienced histologist and pathologist to confirm the technical quality of the HG sections and HR-NB-compatible histomorphology. HG area assessment on H&E-stained sections was performed in QuPath, where segmentation was achieved using pixel classifiers adapted for each image. For descriptive purposes, cell clusters in 3-week H&E sections were additionally segmented using our DANEELpath toolkit for QuPath [47] and validated. Qualitative inspection of both H&E and anti-VN staining was performed at low magnification to assess overall section integrity and cell cluster density/distribution. At higher magnification, cellular organisation and peri-cluster regions were examined to identify features varying across VN microenvironments and CLG treatment in both staining techniques. These observations guided the selection of descriptors later prioritised for digital pathology quantification. Super-resolution imaging of vinculin and β3 integrin was performed using the ZEISS LSM 980 with Airyscan 2, equipped with a Plan-Apochromat 63x objective and a scan zoom of 1.9. Following image acquisition, the Airyscan 2 deconvolution algorithm was applied using the ZEN Microscopy Software (ZEISS) to each of the samples, resulting in a resolution of 0.0353 μm/pixel for all samples. Vinculin and β3 integrin signals appeared as bright punctate structures, but background fluorescence was present due to the FFPE processing. Background removal was performed with a custom pipeline using FIJI [48] and Python. Briefly, vinculin and β3 channels were isolated using a custom FIJI macro. Next, a Python pipeline was generated by using pyClesperanto package [49]. The process involved applying a Difference-of-Gaussians (DoG) filter (σ = 2 vs. σ = 10). The resulting DoG image was duplicated, and a sum of oriented openings was applied using linear kernels (20 pixels long) at angular increments of 5° (from 0° to 180°). Finally, the sum of oriented openings was subtracted from the DoG image, enhancing the dot-like structures characteristic of vinculin and β3 integrin. A second custom FIJI macro was used to generate a composite image combining DAPI, the original vinculin/β3 channel, and the enhanced channels. Composites were visually inspected in QuPath as in brightfield images. For one representative figure, vinculin and β3 integrin signals were segmented using QuPath's built-in watershed detection to aid visualization. Next Generation Sequencing of hydrogels We applied Next Generation Sequencing (NGS) panel to discern the effect of HG composition, culture time and CLG treatment on tumour genomics in the four NB cell lines. DNA was extracted from the cells cultured in the HGs using the Maxwell(R) RSC FFPE Plus DNA Kit (Promega, Madison, Wisconsin, U.S.A.) following the manufacturer's instructions. The previously described customized NGS mechanopanel was applied to identify copy number variations (CNVs), small insertions and deletions (indels), and single nucleotide variants (SNVs) [38]. We employed the SeqCap EZ HyperPlus kit (Roche Diagnostics, Basel, Switzerland) and the NextSeq 550 sequencer (Illumina, San Diego, California, U.S.A.) to obtain at least 100x coverage. Raw data were filtered for quality with PRINSEQ-lite v0.20.4 and mapped against human genome GRCh38 with BWA-MEM v0.7.17. Variant calling was performed with a combination of GATK v4.2.0.0.0 and LoFreq v2.1.3.1. of SNV and indel annotation was performed with Variant Effect Predictor software and the Ensembl v105 database. CNV detection was performed using CopywriteR v2.29.0 and processed with CGHcall v2.58.0. Isolation of hydrogel cells, orthotopic inoculation and in vivo imaging To assess whether cells grown within the HGs could be inoculated orthotopically, we developed a pipeline for cell isolation. Cells were transfected with a lentiviral FLUC vector expressing luciferase and mCherry [50]. Transduced cells from all cell lines were cultured in non-VN-added HGs for three weeks. HGs were transferred to six well plates and washed with dPBS. To disrupt the silk fibroin structure, HGs were incubated with Liberase (0.05 mg mL, Liberase HT, Merck) for 10 min at 37 °C, followed by gentle mechanical disaggregation and an additional 10-min incubation. Liberase activity was reduced by adding FBS to a final concentration of 10%. Cells were centrifuged to remove Liberase and resuspended in culture medium. A second centrifugation step was followed by trypsinization for 5 min at 37 °C to achieve cell individualisation. Trypsin activity was stopped by adding 10% FBS, and viable cells were counted. The cell pellet was washed and resuspended in dPBS. For orthotopic inoculation, Matrigel (Corning, 356234) was mixed with the cell suspension in dPBS at a 1:1 ratio, and one million cells were injected into the left adrenal gland of a total of 52 RAG1−/−/VN−/− mice (RAG1−/− B6.129S7 Rag1tm1Mom J; VN−/− B6.129S2 Vtntm1Dgi J) [28]. Animals were maintained under specific pathogen-free conditions with controlled 12 h light/dark cycles and ad libitum access to food and water. Tumour progression was monitored using bioluminescence (luciferase) assay every 10 days at the University of Valencia Animal Care core facilities. Briefly, 10 min before image capture mice were submitted to intraperitoneal injection of 150 mg/kg body weight of luciferin (D-Luciferin potassium salt, AAT Bioquest). Induction with isoflurane and capture of bioluminescence was achieved using the IVIS system (IVIS Lumina X5 Imaging System). If no tumour was detected after two months, assay frequency was reduced to once per month. Once tumour presence was confirmed, animals were enrolled in a parallel experimental protocol involving repeated intraperitoneal administrations of either CLG (n = 24) or vehicle control (n = 28). For animal experimentation CLG was dissolved in saline solution and administered at a dose of 20 mg/kg body weight every two days for a total of six doses, as previously described [31]. Control animals received six saline injections. At the experimental endpoint, animals were euthanised by cervical dislocation. When macroscopically detectable, the left adrenal gland with tumour mass was collected with other organs and fixed in FFPE for histological analysis. Approval for experiments and ethics declarations We used PDX cell lines from patients. This study was approved by the Research Ethics Committee of the Hospital Clínico Universitario de València (No. 2020/025, Act: 372, 30 September 2021). All experiments were performed in accordance with the guidelines and regulations. Participants or their family members/legal guardians provided written informed consent for the research studies performed in our laboratory. All procedures involving animals were performed in accordance with the approved protocol of the Ethics Committee of the University of València (code 2021/VSC/PEA/0052). Results GTA–sf fabrication protocol enables HG polymerisation and cell embedding Using the GTA–sf enzymatic crosslinking workflow, we consistently obtained fully polymerised HGs with successful cell embedding across all four HR-NB models. A schematic overview of the workflow is provided in Fig. 1 to facilitate adoption and standardisation by subsequent users.Fig. 1HG GTA-sf crafting protocol A) Weigh solid GTA (i) and perform UV sterilization (ii). Melt GTA with cell culture media (iii) and mix with sf (iv). B) Cell trypsinization (i-iii), viability count (iv), centrifugation (v) and resuspension with GTA-sf mix (vi). C) Add VN/PBS and HRP in an eppendorf tube and homogenize with 50 μL of GTA-sf and cells. Place 2 μL of H2O2 at the centre of a 24 well plate. Mix the eppendorf content with H2O2 and let the mix stand for 45 min at 37 °C to trigger enzymatic polymerisation. HG GTA-sf crafting protocol A) Weigh solid GTA (i) and perform UV sterilization (ii). Melt GTA with cell culture media (iii) and mix with sf (iv). B) Cell trypsinization (i-iii), viability count (iv), centrifugation (v) and resuspension with GTA-sf mix (vi). C) Add VN/PBS and HRP in an eppendorf tube and homogenize with 50 μL of GTA-sf and cells. Place 2 μL of H2O2 at the centre of a 24 well plate. Mix the eppendorf content with H2O2 and let the mix stand for 45 min at 37 °C to trigger enzymatic polymerisation. The elastic modulus of VN-rich 3D models tends to be higher than non-VN-added matrices Mechanical properties of GTA–sf HGs were assessed by elastic modulus measurements after two and three weeks of culture under non-VN-added and VN-rich conditions (Fig. 2). Force–deformation curves (Supplementary Materials 1) revealed median elastic modulus values ranging from 0.71 to 2.55 kPa across all conditions, falling within the 0.17-8.45 kPa range previously reported for HR-NB tissues [21]. Acellular HGs showed that VN incorporation did not significantly change stiffness, suggesting that VN functions primarily as a biochemical rather than structural component of the hydrogel network. No statistically significant differences were detected either within individual cell lines when comparing VN-rich and non-VN-added conditions or at two versus three weeks of culture. Despite this, consistent trends towards increased stiffness on VN-rich matrices were observed. When comparing VN conditions, VN-rich matrices generally exhibited higher elastic moduli than non-VN-added HGs at the 2-week time point in all cell lines, suggesting an early stiffening effect associated with VN incorporation. Although the temporal differences were not statistically significant, a trend toward higher stiffness at early time points and slightly lower values at later stages was observed. Early mechanical increases have previously been reported in enzymatically crosslinked silk-based hydrogels, likely reflecting ongoing post-gelation structural maturation [51]. In parallel, the known susceptibility of these HGs to enzymatic degradation suggests that the later reduction in stiffness may be consistent with progressive matrix remodelling by the embedded cells. Overall, these results indicate that VN incorporation modulates the mechanical properties of GTA–sf HGs in a time- and model-dependent manner, while maintaining stiffness values within physiologically relevant ranges. Including earlier time points such as day 3 and day 7 may provide additional insight into the initial dynamics of matrix remodelling and early VN-driven stiffening which are not fully captured at later stages.Fig. 2Elastic moduli of GTA-sf 3D models across time points and different conditions: A) Box plots with individual data points showing the elastic modulus (kPa) of SK-N-BE(2) 3D HGs at d03, d07, d14, and d24 under non-VN-added and VN-rich conditions. Each point represents an individual measurement. No statistically significant temporal differences were detected; however, the data suggested a trend toward higher elastic modulus at earlier time points and slightly lower values at later stages. B) Box plots of 3D hydrogels without cells, with commercial cell lines, and with PDX models after two and three weeks of culture. Each point represents an individual measurement. At two weeks, VN-rich matrices exhibited higher elastic moduli than their non-VN-added counterparts across all cell lines. This trend of increased stiffness in VN-rich conditions persisted at three weeks in commercial cell lines. A general decrease in stiffness was observed at three weeks, particularly in the cell-free and VN-rich conditions of PDX models. No statistically significant differences were detected between pairwise comparisons using the Wilcoxon signed-rank test. For consistency of data presentation, the d14 and d24 SK-N-BE(2) measurements shown in panel A correspond to the same datasets displayed as 2 weeks and 3 weeks, respectively, in panel B. Elastic moduli of GTA-sf 3D models across time points and different conditions: A) Box plots with individual data points showing the elastic modulus (kPa) of SK-N-BE(2) 3D HGs at d03, d07, d14, and d24 under non-VN-added and VN-rich conditions. Each point represents an individual measurement. No statistically significant temporal differences were detected; however, the data suggested a trend toward higher elastic modulus at earlier time points and slightly lower values at later stages. B) Box plots of 3D hydrogels without cells, with commercial cell lines, and with PDX models after two and three weeks of culture. Each point represents an individual measurement. At two weeks, VN-rich matrices exhibited higher elastic moduli than their non-VN-added counterparts across all cell lines. This trend of increased stiffness in VN-rich conditions persisted at three weeks in commercial cell lines. A general decrease in stiffness was observed at three weeks, particularly in the cell-free and VN-rich conditions of PDX models. No statistically significant differences were detected between pairwise comparisons using the Wilcoxon signed-rank test. For consistency of data presentation, the d14 and d24 SK-N-BE(2) measurements shown in panel A correspond to the same datasets displayed as 2 weeks and 3 weeks, respectively, in panel B. GTA-sf models are suitable for FFPE and histological analysis GTA–sf HGs were compatible with FFPE processing, multiple histological staining techniques, and digital slide acquisition (Fig. 3). H&E staining showed cells organised into well-defined multicellular groupings called cell clusters. As shown in Fig. 3A, clusters within a single slide were distributed throughout the matrix, with morphologies resembling those observed in some HR-NB samples.Fig. 3Histological techniques in 3D models. Images are representative and intended to illustrate the technical capabilities of the GTA–sf hydrogel platform rather than comparative analyses; non-VN-added and VN-rich models are compared in subsequent figures. A) H&E staining of 3D HG sections and human HR-NB. i) Low magnification of a HG section. Scale bar 1 mm. ii) High magnification, cells grow as cell clusters within the HG. Scale bar 100 μm. iii) H&E-stained sample of a HR-NB tumour shows similar cluster-like arrangement. Scale bar 100 μm. B) Immunohistochemical staining of Ki67, Bcl2, Caspase 3, and YAP. Representative images of the SK-N-BE(2) cell line cultured in VN-rich HGs, showing control samples (top row, blue) and CLG-treated samples (bottom row, red). Panels show: i) Ki67, proliferation marker; ii) BCL2, anti-apoptotic marker; iii) Caspase 3, apoptotic marker; iv) YAP. Qualitative examination suggests CLG-treated samples showed reduced Ki67 and BCL2 staining, together with increased Caspase 3 signal. Caspase 3 staining in CLG-treated samples tended to localize preferentially at the cluster periphery. YAP signal is low in both conditions. Scale bars: 20 μm. C) Anti-VN immunohistochemistry of 3D hydrogel sections. i) Schema of the two conditions: in non-VN-added models, VN expression comes from cellular expression (represented as red dots). In VN-rich models, signal includes cell-derived (red dots) and matrix-incorporated VN (orange dots) Schema adapted from Ref. [31]. ii) Low-magnification images. Non-VN-added samples displayed DAB signal at cell clusters; VN-rich samples exhibited DAB staining throughout the HG. Scale bar 1 mm. iii) High magnification of clusters, non-VN-added showed intra-cluster VN and peri-cluster expression (green arrow), the VN-corona. VN-rich samples, VN staining present in the matrix and intra-cluster regions, a faint stained region was located at peri-cluster (green arrow), the pale-halo. Scale bar 50 μm. D) Immunofluorescence of neuroblasts (blue, DAPI) vinculin (red) and integrin β3 (green) related to clusters. i) shows the raw image obtained from the super-resolution microscope. ii) After Difference-of-Gaussians filtering and suppression of oriented openings, background is reduced and puncta are enhanced, mainly at pericellular and intercellular levels. Scale bar cell cluster 20 μm. Scale bar inset examination: 3 μm; IHC: immunohistochemistry. Histological techniques in 3D models. Images are representative and intended to illustrate the technical capabilities of the GTA–sf hydrogel platform rather than comparative analyses; non-VN-added and VN-rich models are compared in subsequent figures. A) H&E staining of 3D HG sections and human HR-NB. i) Low magnification of a HG section. Scale bar 1 mm. ii) High magnification, cells grow as cell clusters within the HG. Scale bar 100 μm. iii) H&E-stained sample of a HR-NB tumour shows similar cluster-like arrangement. Scale bar 100 μm. B) Immunohistochemical staining of Ki67, Bcl2, Caspase 3, and YAP. Representative images of the SK-N-BE(2) cell line cultured in VN-rich HGs, showing control samples (top row, blue) and CLG-treated samples (bottom row, red). Panels show: i) Ki67, proliferation marker; ii) BCL2, anti-apoptotic marker; iii) Caspase 3, apoptotic marker; iv) YAP. Qualitative examination suggests CLG-treated samples showed reduced Ki67 and BCL2 staining, together with increased Caspase 3 signal. Caspase 3 staining in CLG-treated samples tended to localize preferentially at the cluster periphery. YAP signal is low in both conditions. Scale bars: 20 μm. C) Anti-VN immunohistochemistry of 3D hydrogel sections. i) Schema of the two conditions: in non-VN-added models, VN expression comes from cellular expression (represented as red dots). In VN-rich models, signal includes cell-derived (red dots) and matrix-incorporated VN (orange dots) Schema adapted from Ref. [31]. ii) Low-magnification images. Non-VN-added samples displayed DAB signal at cell clusters; VN-rich samples exhibited DAB staining throughout the HG. Scale bar 1 mm. iii) High magnification of clusters, non-VN-added showed intra-cluster VN and peri-cluster expression (green arrow), the VN-corona. VN-rich samples, VN staining present in the matrix and intra-cluster regions, a faint stained region was located at peri-cluster (green arrow), the pale-halo. Scale bar 50 μm. D) Immunofluorescence of neuroblasts (blue, DAPI) vinculin (red) and integrin β3 (green) related to clusters. i) shows the raw image obtained from the super-resolution microscope. ii) After Difference-of-Gaussians filtering and suppression of oriented openings, background is reduced and puncta are enhanced, mainly at pericellular and intercellular levels. Scale bar cell cluster 20 μm. Scale bar inset examination: 3 μm; IHC: immunohistochemistry. In addition, GTA–sf HGs were compatible with immunohistochemical staining of Ki67, BCL2, Caspase 3, and YAP (Fig. 3B). In representative SK-N-BE(2) VN-rich HGs, qualitative evaluation suggest CLG-treated samples showed reduced Ki67 and BCL2 staining together with increased Caspase 3 signal compared with controls. Caspase 3 positivity in CLG-treated samples also tended to localize at the cluster periphery. In contrast, YAP signal was low in both conditions. Anti-VN staining revealed a strong DAB signal throughout the matrix in VN-rich conditions, indicating effective incorporation of exogenous VN (Fig. 3C). Within and inside cell clusters, VN was also similarly expressed under both matrix conditions, as we have previously described in malignant neuroblastic cell niches in human biopsies. Notably, anti-VN staining revealed two distinctive peri-cell cluster features. In non-VN-added conditions, some cell clusters were surrounded by an area of VN deposition, referred to as a VN-corona (Fig. 3C). In VN-rich conditions, a lightly stained region was observed around certain cell clusters, which we call a pale-halo. In addition, FFPE sections supported fluorescence staining and super-resolution imaging. Super-resolution microscopy combined with digital image post-processing enabled visualization of adhesion-related proteins, vinculin and integrin β3 at subcellular resolution (Fig. 3D). Qualitative histopathology reveals model-specific morphological and spatial responses to VN microenvironments and CLG treatment H&E staining showed that all four HR-NB models formed multicellular structures within both VN microenvironments and treatment conditions at the 3-week time point. Section area was quantified in QuPath across 44 HG slides cultured for three weeks, yielding a mean ± SD of 11.84 ± 4.52 mm2. For descriptive purposes, DANEELpath-based segmentation of the same H&E sections identified a total of 35,895 detectable cellular structures (including cell clusters and individual cells), corresponding to a mean ± SD of 815.79 ± 667.40 structures per HG. These quantitative descriptors provide a basis for systematic morphometric analysis, including cluster size, shape, density, and spatial distribution, enabling objective comparison across experimental conditions. Such analyses are essential to move beyond qualitative assessment and establish reproducible morphology-based response metrics, and represent a key direction for future studies using this platform. Qualitative evaluation revealed several changes in spatial and morphological features of cell clusters according to VN and CLG environments. Under non-VN-added conditions, SK-N-BE(2) showed the highest cell cluster density among models, and under CLG treatment clusters frequently appeared partially detached from the surrounding matrix; in PDX1, cluster organisation appeared markedly disrupted (Fig. 4A–Di–iv). Under VN-rich conditions, cluster density increased relative to non-VN-added matrices, and larger clusters were more frequent; following CLG treatment, density appeared reduced and clusters tended to be more widely spaced (Fig. 5A–Di–iv). At higher magnification, intercellular spacing within clusters appeared greater under CLG treatment in both VN microenvironments.Fig. 4H&E staining of non-VN-added cultured HGs. H&E staining of 3D model sections cultured in non-VN-added conditions for three weeks. Representative images are shown for the four cell lines after three weeks of culture: A) SK-N-BE (2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. In CLG-treated samples, cells often appear more detached from the surrounding matrix. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area and cells for the cell clusters marked with an asterisk on high augment panels with the following format (area μm2/cell number in Control vs area μm2/cell number in CLG): SK-N-BE(2) (1807.43 μm2/46 vs 2333.75 μm2/42); SH-SY5Y (3631.13 μm2/40 vs 3345.43 μm2/53); PDX1 (6583.82 μm2/76 vs 2322.10 μm2/12); PDX2 (5147.17 μm2/66 vs 6082.05 μm2/73).Fig. 5H&E staining of VN-rich cultured HGs. H&E staining of 3D models sections cultured in VN-rich conditions for three weeks. Representative images are shown for the four cell lines after three weeks of culture: A) SK-N-BE(2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. VN-rich conditions showed more cell cluster density. Following CLG treatment, clusters were more widely spaced and rounded. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area and cells for the cell clusters marked with an asterisk on high augment panels with the following format (area μm2/cell number in Control vs area μm2/cell number in CLG): SK-N-BE(2) (1803.75μm2/32 vs 1282.64 μm2/28); SH-SY5Y (2341.71 μm2/32 vs 3130.07 μm2/46); PDX1 (5219.04 μm2/45 vs 5929.32μm2/62); PDX2 (7487.87 μm2/93 vs 5102.27 μm2/80). H&E staining of non-VN-added cultured HGs. H&E staining of 3D model sections cultured in non-VN-added conditions for three weeks. Representative images are shown for the four cell lines after three weeks of culture: A) SK-N-BE (2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. In CLG-treated samples, cells often appear more detached from the surrounding matrix. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area and cells for the cell clusters marked with an asterisk on high augment panels with the following format (area μm2/cell number in Control vs area μm2/cell number in CLG): SK-N-BE(2) (1807.43 μm2/46 vs 2333.75 μm2/42); SH-SY5Y (3631.13 μm2/40 vs 3345.43 μm2/53); PDX1 (6583.82 μm2/76 vs 2322.10 μm2/12); PDX2 (5147.17 μm2/66 vs 6082.05 μm2/73). H&E staining of VN-rich cultured HGs. H&E staining of 3D models sections cultured in VN-rich conditions for three weeks. Representative images are shown for the four cell lines after three weeks of culture: A) SK-N-BE(2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. VN-rich conditions showed more cell cluster density. Following CLG treatment, clusters were more widely spaced and rounded. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area and cells for the cell clusters marked with an asterisk on high augment panels with the following format (area μm2/cell number in Control vs area μm2/cell number in CLG): SK-N-BE(2) (1803.75μm2/32 vs 1282.64 μm2/28); SH-SY5Y (2341.71 μm2/32 vs 3130.07 μm2/46); PDX1 (5219.04 μm2/45 vs 5929.32μm2/62); PDX2 (7487.87 μm2/93 vs 5102.27 μm2/80). Anti-VN staining enabled qualitative assessment of intra-cluster VN expression and peri-cluster features across all HR-NB models in the HG platform under both VN microenvironments. Under non-VN-added conditions, VN staining was detected within cell clusters across all models. Notably, the peri-cluster VN-corona was observed in the MYCN-amplified cells SK-N-BE(2), PDX1 and PDX2. Following CLG treatment, VN-corona structures were reduced in SK-N-BE(2) and PDX1, while PDX2 showed heterogeneous responses, with some clusters retaining VN-corona and others lacking it (Fig. 6A–D.i–iv). Under VN-rich conditions, intra-cluster VN differed between commercial and PDX models, with stronger intra-cluster VN signal in PDX lines. At the peri-cluster level, pale-halo was observed, particularly prominent in SH-SY5Y and SK-N-BE(2), and this feature was reduced following CLG treatment (Fig. 7A–D.i–iv).Fig. 6Anti-VN staining of non-VN-added cultured HGs. Anti-VN staining of 3D model sections cultured in non-VN-added conditions for three weeks. Representative images for the four cell lines: A) SK-N-BE(2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. VN staining is localized within cell clusters, and the peri-cluster structure VN-corona is visible in the MYCN-amplified cell lines SK-N-BE(2), PDX1, and PDX2. VN-corona was generally reduced following CLG treatment. In PDX2, CLG-treated samples showed heterogeneity, with some clusters retaining the VN-corona and others lacking it. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area intra-cluster VN proportion (area Anti-VN/area cluster), and VN-corona width [(radius cluster + corona) – (radius cluster)] marked with an asterisk on high augment panels with the following format (area μm2/intracluster VN/VN-corona μm in Control vs area μm2/intracluster VN/VN-corona μm in CLG): SK-N-BE(2) (1642.23 μm2/0.11/8.38 μm vs 3169.29μm2/0.38/0.1 μm); SH-SY5Y (2327.04 μm2/0.09/0 μm vs 2023.18 μm2/0.68/0 μm); PDX1 (8627.12 μm2/0.13/4.22 μm vs 1228.52 μm2/0.02/0 μm); PDX2 (4296.33 μm2/0.23/24.4 μm vs 14231.13 μm2/0.08/18.66 μm).Fig. 7Anti-VN staining of VN-rich cultured HGs Anti-VN staining of 3D model sections cultured in VN-rich conditions for three weeks. Representative images are shown for the four cell lines: A) SK-N-BE(2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. PDX cell lines showed higher levels of intra-cluster VN expression compared to commercial cell lines. In non-treated samples, we observed peri-cluster pale-halo, particularly prominent in SH-SY5Y and SK-N-BE(2). This pale-halo was reduced following CLG treatment. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area intra-cluster VN proportion(area Anti-VN/area cluster), and pale-halo width ((radius cluster + pale-halo) – (radius cluster)) marked with an asterisk on high augment panels with the following format (area μm2/intracluster VN/VN-corona μm in Control vs area μm2/intracluster VN/VN-corona μm in CLG): SK-N-BE(2) (1451.24 μm2/0.02/4.61 μm vs 4172.28μm2/0.13/4.08 μm); SH-SY5Y (2677.06μm2/0.08/9.37 μm vs 5919.5266μm2/0.03/1.98 μm); PDX1 (6371.04 μm2/0.55/2.36 μm vs 4194.23 μm2/0.25/0 μm); PDX2 (4327.32 μm2/0.40/10.57 μm vs 5406.51 μm2/0.17/0.99 μm). Anti-VN staining of non-VN-added cultured HGs. Anti-VN staining of 3D model sections cultured in non-VN-added conditions for three weeks. Representative images for the four cell lines: A) SK-N-BE(2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. VN staining is localized within cell clusters, and the peri-cluster structure VN-corona is visible in the MYCN-amplified cell lines SK-N-BE(2), PDX1, and PDX2. VN-corona was generally reduced following CLG treatment. In PDX2, CLG-treated samples showed heterogeneity, with some clusters retaining the VN-corona and others lacking it. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area intra-cluster VN proportion (area Anti-VN/area cluster), and VN-corona width [(radius cluster + corona) – (radius cluster)] marked with an asterisk on high augment panels with the following format (area μm2/intracluster VN/VN-corona μm in Control vs area μm2/intracluster VN/VN-corona μm in CLG): SK-N-BE(2) (1642.23 μm2/0.11/8.38 μm vs 3169.29μm2/0.38/0.1 μm); SH-SY5Y (2327.04 μm2/0.09/0 μm vs 2023.18 μm2/0.68/0 μm); PDX1 (8627.12 μm2/0.13/4.22 μm vs 1228.52 μm2/0.02/0 μm); PDX2 (4296.33 μm2/0.23/24.4 μm vs 14231.13 μm2/0.08/18.66 μm). Anti-VN staining of VN-rich cultured HGs Anti-VN staining of 3D model sections cultured in VN-rich conditions for three weeks. Representative images are shown for the four cell lines: A) SK-N-BE(2), B) SH-SY5Y, C) PDX1, and D) PDX2. For each cell line, control samples (blue column heading) are displayed in columns i) and ii), corresponding to low and high magnification, respectively. Treated samples (red column heading), shown in columns iii) and iv), received six cumulative doses of CLG at 100 μM. Green squares in low-magnification panels indicate the areas shown at higher magnification. PDX cell lines showed higher levels of intra-cluster VN expression compared to commercial cell lines. In non-treated samples, we observed peri-cluster pale-halo, particularly prominent in SH-SY5Y and SK-N-BE(2). This pale-halo was reduced following CLG treatment. Scale bars: 100 μm for low magnification, 20 μm for high magnification. Below are the measurements for cell cluster area intra-cluster VN proportion(area Anti-VN/area cluster), and pale-halo width ((radius cluster + pale-halo) – (radius cluster)) marked with an asterisk on high augment panels with the following format (area μm2/intracluster VN/VN-corona μm in Control vs area μm2/intracluster VN/VN-corona μm in CLG): SK-N-BE(2) (1451.24 μm2/0.02/4.61 μm vs 4172.28μm2/0.13/4.08 μm); SH-SY5Y (2677.06μm2/0.08/9.37 μm vs 5919.5266μm2/0.03/1.98 μm); PDX1 (6371.04 μm2/0.55/2.36 μm vs 4194.23 μm2/0.25/0 μm); PDX2 (4327.32 μm2/0.40/10.57 μm vs 5406.51 μm2/0.17/0.99 μm). Super-resolution imaging at the 3-week time point, combined with digital post-processing and segmentation, enabled qualitative assessment of vinculin and β3 integrin localisation in the HGs (Fig. 8A–D.i–iv). Despite the loss of some VN-rich PDX TMA cores, qualitative analysis was conducted. Under control conditions, vinculin was detected within cell clusters across all models, with additional peri-cluster signal in SK-N-BE(2) and in non-VN-added PDX samples. β3 signal was enriched at the cluster periphery in SK-N-BE(2) and the PDX models, with peri-cluster localisation most evident in SK-N-BE(2) and less prominent in the PDX samples, whereas SH-SY5Y showed a more homogeneous intra-cluster distribution. Following CLG treatment, SK-N-BE(2) showed a shift of vinculin towards the cluster centre with reduced peri-cluster staining, and β3 signal decreased in SK-N-BE(2) and SH-SY5Y.Fig. 8Super-resolution microscopy of vinculin and β3 integrin in big cell clusters from HR-NB models. Merged images of segmented vinculin (magenta), segmented β3 integrin (yellow), and DAPI (blue) are shown for commercial cell lines: (A) SK-N-BE(2) and (B) SH-SY5Y, and PDX models: (C) PDX1 and (D) PDX2. Columns represent the experimental conditions: (i) non-VN-added Control, (ii) non-VN-added CLG, (iii) VN-rich Control, and (iv) VN-rich CLG. VN-rich conditions were not available for PDX models, and PDX2 non-VN-added Control was not available due to technical issues. Vinculin was detected within clusters across all models, with peri-cluster signal most evident in SK-N-BE(2) and less prominent in non-VN-added PDX samples. β3 integrin was enriched at the cluster periphery in SK-N-BE(2) and PDX models, with peri-cluster localisation more frequent in SK-N-BE(2), whereas SH-SY5Y showed a more homogeneous intra-cluster pattern. After CLG, SK-N-BE(2) showed a central shift of vinculin with reduced peri-cluster localisation, and β3 signal decreased in SK-N-BE(2) and SH-SY5Y. Scale bar: 20 μm. Below are the measurements for the number of vinculin and β3 integrin detections in the cell clusters shown in the figure, in the following format (number vinculin/number beta3). SK-N-BE(2): (1560/550 non-VN-added-control; 1123/279 non-VN-added-CLG; 1127/677 VN-rich-control; 697/366 VN-rich-CLG): ; SH-SY5Y (1729/1534 non-VN-added-control; 1009/1712 non-VN-added-CLG; 2122/1301 VN-rich-control; 1516/465 VN-rich-CLG); PDX1, non-VN-added only, (5258/3475, Control; 9725/6639 in CLG); PDX2, non-VN-added only (6225/3006, Control; 3458/1837 in CLG). Super-resolution microscopy of vinculin and β3 integrin in big cell clusters from HR-NB models. Merged images of segmented vinculin (magenta), segmented β3 integrin (yellow), and DAPI (blue) are shown for commercial cell lines: (A) SK-N-BE(2) and (B) SH-SY5Y, and PDX models: (C) PDX1 and (D) PDX2. Columns represent the experimental conditions: (i) non-VN-added Control, (ii) non-VN-added CLG, (iii) VN-rich Control, and (iv) VN-rich CLG. VN-rich conditions were not available for PDX models, and PDX2 non-VN-added Control was not available due to technical issues. Vinculin was detected within clusters across all models, with peri-cluster signal most evident in SK-N-BE(2) and less prominent in non-VN-added PDX samples. β3 integrin was enriched at the cluster periphery in SK-N-BE(2) and PDX models, with peri-cluster localisation more frequent in SK-N-BE(2), whereas SH-SY5Y showed a more homogeneous intra-cluster pattern. After CLG, SK-N-BE(2) showed a central shift of vinculin with reduced peri-cluster localisation, and β3 signal decreased in SK-N-BE(2) and SH-SY5Y. Scale bar: 20 μm. Below are the measurements for the number of vinculin and β3 integrin detections in the cell clusters shown in the figure, in the following format (number vinculin/number beta3). SK-N-BE(2): (1560/550 non-VN-added-control; 1123/279 non-VN-added-CLG; 1127/677 VN-rich-control; 697/366 VN-rich-CLG): ; SH-SY5Y (1729/1534 non-VN-added-control; 1009/1712 non-VN-added-CLG; 2122/1301 VN-rich-control; 1516/465 VN-rich-CLG); PDX1, non-VN-added only, (5258/3475, Control; 9725/6639 in CLG); PDX2, non-VN-added only (6225/3006, Control; 3458/1837 in CLG). Altogether, the GTA–sf HG platform enabled qualitative identification of diverse model-specific morphological and spatial features across VN microenvironments and CLG treatment. Cells cultured on 3D models retained the genotype of 2D cell lines To explore potential culture-driven genomic changes we assessed the genetic profiles of all four NB models. NGS analysis showed minimal changes in CNVs across all four NB cell lines, including culture duration, VN incorporation, and CLG treatment, as shown in Fig. 9.Fig. 9Chromosomal aberrations in the 3D HGs found by Next Generation Sequencing. Genetic alterations were assessed in all four cell lines after two and three weeks of 3D culture. Solid blue and red rectangles indicate chromosomal gains and losses, respectively, for the corresponding chromosome arms. Asterisks denote regions with insufficient probe coverage to confirm alterations. Common alterations associated with HR-NB were detected, including +2p in SK-N-BE(2) and PDX1, −1p in MYCN-amplified cell lines, +1q in SH-SY5Y, and +17q in both PDX models and SH-SY5Y. Additionally, in SK-N-BE(2) hydrogels cultured for three weeks with added VN, a deletion was observed on chromosome 9q-(24.2-22.33). In SH-SY5Y, a loss of +2p (ter–16.3) was detected; however, this alteration was not homogeneous in 2D as it was present in only 31% of cells as indicated by the empty blue. All other alterations were previously identified in 2D cultures or had been reported earlier. Chromosomal aberrations in the 3D HGs found by Next Generation Sequencing. Genetic alterations were assessed in all four cell lines after two and three weeks of 3D culture. Solid blue and red rectangles indicate chromosomal gains and losses, respectively, for the corresponding chromosome arms. Asterisks denote regions with insufficient probe coverage to confirm alterations. Common alterations associated with HR-NB were detected, including +2p in SK-N-BE(2) and PDX1, −1p in MYCN-amplified cell lines, +1q in SH-SY5Y, and +17q in both PDX models and SH-SY5Y. Additionally, in SK-N-BE(2) hydrogels cultured for three weeks with added VN, a deletion was observed on chromosome 9q-(24.2-22.33). In SH-SY5Y, a loss of +2p (ter–16.3) was detected; however, this alteration was not homogeneous in 2D as it was present in only 31% of cells as indicated by the empty blue. All other alterations were previously identified in 2D cultures or had been reported earlier. Common HR-NB characteristics known to be present in the 2D cell cultures were preserved in the 3D models, including 2p+ in SK-N-BE (2) and PDX-1, where MYCN oncogene is located. Other alterations such as 1p-were consistent in all conditions of MYCN-amplified cell lines, whereas 1q+ was observed in all SH-SY5Y conditions. Another common HR-NB alteration maintained was 17q+, where the VN gene is located, in PDX models and SH-SY5Y. The main changes observed, in SK-N-BE (2) untreated HGs at three weeks with added VN, showed 9q-(24.2-22.33). In the SH-SY5Y cell line the only change detected was the disappearance of +2p(ter-16.3), although this alteration was present in approximately 31% of cells in the 2D cultures, suggesting no major shift in the overall profile. Our study of SNVs or indels revealed no new variants or changes in allele frequencies in any of the four NB cell lines when cultured in HGs. Only those already found in the 2D cultures were detected, notably the pathogenic mutations p.C135F in TP53 of SK-N-BE (2), p.F1174L in ALK and p.G12V in KRAS of SH-SY5Y. Cells isolated from HGs and orthotopically injected into mice formed neuroblastic tumours To assess the downstream in vivo applicability of the GTA–sf HG platform, a workflow was established to isolate HR-NB cells from 3-week cell culture and prepare them for orthotopic inoculation. As shown in Fig. 10A i-iii, HR-NB cells were successfully transfected with luciferase-mCherry plasmids and embedded into the HGs. Cell cluster isolation and subsequent cell individualisation were successfully achieved, as shown in Fig. 10B i-iii.Fig. 10Cell cluster isolation and in vivo orthotopic inoculation. A) Plasmid transfection in HG grown cells: i) Phase-contrast image of cells cultured for three weeks in HGs; ii) mCherry fluorescence signal (magenta); iii) overlay of phase-contrast and fluorescence channels. Scale bar: 200 μm. B) Isolation of cell clusters from HGs: i) shows clusters embedded within the HG matrix; ii) shows cluster; iii) shows individual cells obtained by subsequent trypsinization. Scale bar: 100 μm. C) In vivo bioluminescence imaging of orthotopic inoculations in the left adrenal gland. SK-N-BE(2) and SH-SY5Y showed luciferase signal 10 days after inoculation, whereas PDX1 and PDX2 models exhibited detectable luciferase after 59 and 246 days, respectively. Cell cluster isolation and in vivo orthotopic inoculation. A) Plasmid transfection in HG grown cells: i) Phase-contrast image of cells cultured for three weeks in HGs; ii) mCherry fluorescence signal (magenta); iii) overlay of phase-contrast and fluorescence channels. Scale bar: 200 μm. B) Isolation of cell clusters from HGs: i) shows clusters embedded within the HG matrix; ii) shows cluster; iii) shows individual cells obtained by subsequent trypsinization. Scale bar: 100 μm. C) In vivo bioluminescence imaging of orthotopic inoculations in the left adrenal gland. SK-N-BE(2) and SH-SY5Y showed luciferase signal 10 days after inoculation, whereas PDX1 and PDX2 models exhibited detectable luciferase after 59 and 246 days, respectively. In vivo bioluminescence imaging confirmed orthotopic engraftment of all four models into the left adrenal gland (Fig. 10C). Luciferase signal was detected in tumours derived from the commercial cell lines SK-N-BE(2) and SH-SY5Y as early as 10 days post-inoculation. In contrast, PDX-derived models exhibited slower engraftment kinetics, with detectable bioluminescence signals at 59 days for PDX1 and 246 days for PDX2, respectively. Histopathological evaluation of collected tissues confirmed the presence of orthotopic tumours, which displayed the characteristic undifferentiated small round blue cell morphology of HR-NB (Fig. 11). Engraftment rates differed between genetic backgrounds, with MYCN-amplified cell lines (SK-N-BE(2), PDX1 and PDX2) showing higher tumour engraftment than the ALK-mutated SH-SY5Y cell line.Fig. 11Histopathological evaluation of orthotopic tumours with H&E staining samples obtained from mice orthotopically injected with SK-N-BE(2), SH-SY5Y, PDX1 and PDX2 cells. A) Low-magnification images show tumour growth adjacent to the kidney. The adrenal gland was not visible. Green squares indicate the region selected for high magnification. Scale bar 2000 μm. B) High-magnification images reveal tumours composed of small, round, blue cells, consistent with the histological appearance of HR-NB. Scale bar 100 μm. Histopathological evaluation of orthotopic tumours with H&E staining samples obtained from mice orthotopically injected with SK-N-BE(2), SH-SY5Y, PDX1 and PDX2 cells. A) Low-magnification images show tumour growth adjacent to the kidney. The adrenal gland was not visible. Green squares indicate the region selected for high magnification. Scale bar 2000 μm. B) High-magnification images reveal tumours composed of small, round, blue cells, consistent with the histological appearance of HR-NB. Scale bar 100 μm. Discussion The use of 3D HGs has been widely adopted in cancer research, with notable applications affording a deeper understanding of the biology of multiple pathologies such as glioblastoma tumour and prostate tumour carcinoma. The range of strategies has broadened from organoid systems to the diverse selection of biomaterials used in these studies, including gelatin methacrylate (GelMA), alginate, hyaluronan, polyethylene glycol (PEG), cellulose-based scaffolds, and Matrigel [22,23,[52], [53], [54]]. While Matrigel-based systems provide a biologically relevant ECM environment, they are complex and incompletely defined basement membrane extracts enriched in laminin and collagen IV, and therefore offer limited control over matrix composition. Similarly, although organoid systems are highly valuable, they often rely on complex and variable extracellular matrices that can limit precise control over individual matrix cues [55]. Importantly, beyond general biomaterial diversity, a key limitation of several synthetic platforms such as PEG-based systems lies in their reduced intrinsic bioactivity and dependence on chemical functionalisation to introduce adhesion ligands [56]. In contrast, the GTA–sf system combines inherent bioactivity with enzymatically driven crosslinking, enabling direct incorporation of ECM proteins such as VN without additional modification steps. This platform therefore enables controlled modulation of VN as a defined adhesion cue in a reproducible and experimentally accessible setting [20]. In HR-NB research, 3D models have been developed using diverse approaches, including bioprinting, microfluidic platforms, and HG cultures of commercial cell lines, PDX, and patient derived organoid (PDO) models and co-culture systems [22,23,[57], [58], [59], [60]]. While these strategies provided valuable insights, most reported systems are limited to short culture durations (typically ≤15 days), which constrains long-term studies and cumulative drug testing. Our group has previously explored several 3D modelling strategies, including GelMA-based bioprinted systems that enable long-term culture of commercial HR-NB cell lines for up to 12 weeks and support downstream genetic and histological analyses. However, these models relied on specialised 3D bioprinting equipment and did not incorporate VN as a matrix component [22]. A PEG–maleimide HG system was also evaluated, allowing controlled stiffness, degradability and VN incorporation via the chemical addition of PEG arms, but it involved complex chemical handling, was studied in shorter culture times (up to 14 days), and was not compatible with standard FFPE histological processing [23]. Thus, whereas the previous VN/PEG platform provided a more chemically defined proof of concept for matrix-bound VN under short-term conditions, the present GTA–sf/VN system offers a complementary platform for longer-term, histology-compatible analysis of VN-dependent phenotypes. In this context, the GTA–sf HG platform offers a simplified and versatile 3D system, enabling VN incorporation via enzymatic dityrosine crosslinking. The biological relevance of the system arises from its ability to combine controlled VN availability with a protein-based 3D environment that supports cell–matrix remodelling. In the present study, VN-rich conditions were associated with measurable differences in cluster organisation and treatment-associated phenotypes relative to non-VN-added conditions, supporting the functional relevance of VN within this matrix setting. Previous research has demonstrated its suitability for advanced analysis such as holographic microscopy and detection of VN secretion by ELISA [32,33]. This earlier work showed no detectable VN in the culture media of cell-free HGs by ELISA, indicating a stable incorporation and retention of VN within the HG matrix [32]. In the present study, we have shown that HGs mechanical properties fall within HR-NB values [21], enable long-term culture of both commercial and PDX cell lines, integrin-targeted drug response, and are fully compatible with FFPE histology and advanced imaging. This range of histological and imaging approaches enabled qualitative examination of hundreds of cell clusters per tissue section. The histological description focused primarily on the 3-week time point, when differences in tissue organisation and cell–matrix interactions are more readily observable across experimental conditions. Nonetheless, the models were successfully analysed at various time points during the cultivation process. The VN-rich matrix platform has enabled greater cluster density and proximity between clusters, as well as increased cell–cell and cell–ECM interactions that may promote proliferation, autocrine chemokine gradients, and collective migratory events, as described in other 3D systems [[61], [62], [63]]. Anti-VN staining demonstrated that VN expression appears in both non-VN-added and VN-rich matrices with variations between cell lines. These peri-cluster structures likely reflect dynamic ECM remodelling and integrin-mediated adhesion processes. From a mechanobiological perspective, these observations can be interpreted within a framework in which tumour cell behaviour arises from the integration of ligand presentation and matrix mechanics. This concept is consistent with established materiomechanobiology principles, where cell fate emerges from the coordinated integration of ECM composition and bulk mechanical cues through multiscale mechanotransduction processes [64]. VN incorporation provides adhesive ligands that can promote αv-integrin engagement, leading to integrin clustering and adhesion reinforcement at the cell-matrix interface. This process is expected to enhance focal adhesion assembly and cytoskeletal tension, thereby activating downstream mechanotransduction pathways such as FAK/Src signalling and RhoA-mediated contractility, with potential convergence on transcriptional regulators including YAP/TAZ. While these observations are consistent with integrin engagement, it is important to note that the present study does not include direct quantification of downstream signalling activity, and these pathways remain to be functionally validated in future work. Importantly, these ligand-driven effects act in concert with bulk matrix stiffness, which modulates force transmission and cellular mechanosensing, ultimately shaping tumour phenotype. The formation of VN-corona and pale-halo regions may indicate localised secretion, degradation, or redistribution of VN driven by tumour cells, consistent with the adaptive adhesion mechanisms and matrix remodelling observed in other tumour systems. Within this context, these features may also reflect spatial he biological relevance of the system arises fromorganisation of adhesion forces and integrin-mediated signalling, emerging from biochemical–mechanical coupled regulation of the microenvironment [65]. Such features further support the relevance of this platform to study spatially resolved cell–ECM interactions and enable the study of specific regulatory mechanisms for each cell line, which may involve feedback mechanisms or differential use of the matrix [[66], [67], [68]]. In addition, we observed distinct peri-cluster VN structures in HR-NB models, like the ones observed on Ewing Sarcoma cells cultured on the same HGs [69]. VN-corona features were mainly associated with MYCN-amplified models, whereas pale-halo structures were most prominent in the ALK-mutated SH-SY5Y, suggesting genotype-dependent responses to VN availability that should be further explored using more cell lines in this platform. Systematic quantification of such features could therefore help identify morphology-based treatment response indicators, in line with morphological profile approaches [70,71]. It is important to highlight that this study focuses primarily on developing the GTA–sf HG platform as a potential predictive and drug-screening platform rather than comprehensive biological endpoint analysis. The morphological features illustrated in Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8, Fig. 9, Fig. 10, Fig. 11 serve as representative examples of platform performance. Notably, these features can be systematically quantified using our DANEELpath toolkit, enabling robust morphometric and spatial analysis of cell clusters and peri-cluster regions. This quantitative capability underscores the platform's potential for future high-content studies of cell–ECM interactions, integrin-dependent signalling, and drug-response profiling. In this context, digital pathology and automation are essential. Existing tools such as BEHAV3D [72] and SpheroidJ [73] are primarily optimised for live imaging, fluorescence microscopy, or single-spheroid segmentation, rather than for FFPE whole-slide analysis of engineered 3D HGs. To address this gap, our group developed the DANEELpath toolkit for QuPath [47] to enable scalable cell cluster segmentation and extract morphological and spatial features from cell clusters, individual cells, and super-resolution images in these specific 3D HG models. Combined 3D HG culture and digital pathology may complement large-scale HR-NB drug platforms [74] by enabling phenotypic assessment of integrin- and matrix-targeted therapies in long-term cell culture, multi-dose settings. The observed spatial and morphological changes following CLG treatment suggest context-dependent integrin inhibition effects that may not be captured in conventional systems. Genetic alterations are central to tumour biology and therapeutic stratification, with both inter- and intra-tumour heterogeneity influencing tumour behaviour and responses to targeted therapies [[75], [76], [77], [78]]. Given that long-term 2D culture and microenvironmental perturbations can promote clonal selection and genetic drift [79,80], preserving the original genetic landscape is essential for meaningful drug testing. In this study, HR-NB cells cultured within GTA–sf HGs retained their core genomic features across VN microenvironments and CLG treatment. Notably, loss of the +2p(ter 16.3) gain in SH-SY5Y has previously been reported between 2D and 3D culture in other HG systems, and the gain was present in only 31% of cells in 2D culture [38]. While our analyses focused on genomic copy number, SNVs, and indels, integrating transcriptomic profiling would provide complementary insights into microenvironment-driven gene expression changes, pathway activation, and cell-state heterogeneity. Such analyses could reveal ECM- and VN-dependent transcriptional programs, identify mechanotransduction-responsive genes, and uncover early molecular indicators of integrin-targeted therapy response that are not apparent from DNA-level analyses. In combination with spatial and morphological features derived from digital pathology, transcriptomics could therefore provide a more comprehensive understanding of how long-term 3D culture and VN-rich microenvironments shape tumour behaviour and therapeutic susceptibility [81,82]. Together, these findings support that GTA-sf HGs maintain core genomic features while enabling the emergence of microenvironment-driven phenotypes, including multicellular organisation, integrin redistribution, and context-dependent responses to VN and CLG, highlighting the importance of decoupling genetic stability from functional tumour behaviour in 3D culture systems. This is particularly relevant for phenotypic drug screening approaches that integrate genetic context, as demonstrated by recent large-scale HR-NB drug screening platforms combining genomics with drug response profiling [74]. Another strength of this GTA–sf HG platform is its ability to recover viable cells after long-term 3D culture for downstream applications, including orthotopic implantation. Orthotopic xenografts are a gold standard in HR-NB and PDX models show variable engraftment kinetics while preserving tumour-specific phenotypes [79,83]. This experimental workflow, integrating tuneable 3D culture with downstream orthotopic implantation of recovered cells, enables a new research framework in which ECM composition and stiffness can be systematically varied in vitro and their effects tracked in vivo. It is expected that sustained microenvironmental conditioning activates mechanotransduction pathways that persist after extraction, as a form of mechanical memory, thereby influencing tumour cells behaviour after implanation. Future studies comparing cells preconditioned in VN-rich versus VN-poor matrices, or previously exposed to Cilengitide, will be important to assess how microenvironmental and pharmacological preconditioning influence engraftment efficiency, tumour growth kinetics, and histological characteristics. Moreover, this recovery can facilitate downstream single-cell analyses to investigate microenvironment-driven adaptation and differentiation trajectories [84]. Future work should incorporate orthotopic transplantation features, quantitative assessment of mechanotransduction and cell fate markers, including YAP/TAZ localisation, proliferation markers such as Ki67, and apoptosis markers such as cleaved caspase-3. Integrating these markers with spatial and morphometric features derived from digital pathology could provide a more comprehensive framework to link ECM mechanics with tumour cell behaviour and therapeutic response. In summary, we present a versatile 3D GTA–sf HG platform with tuneable VN microenvironments, demonstrating that matrix engineering, long-term culture, drug testing, histopathological assays, molecular analysis, and in vivo translation can be integrated within a single 3D system. Its applicability is illustrated in HR-NB, with potential across other childhood cancer models. Funding sources The authors gratefully acknowledge funding support from 10.13039/501100004587ISCIII (FIS) and 10.13039/501100002924FEDER (European Regional Development Fund) grant number PI20/01107; 10.13039/501100014139CIBERONC (contractCB16/12/00484); FNB (Fundación Neuroblastoma) grant number PRV/00166. IVM was supported by the 10.13039/100014440Ministry of Science, Innovation and Universities of Spain (FPU20/05344). CRediT authorship contribution statement Isaac Vieco-Martí: Conceptualization, Investigation, Methodology, Software, Validation, Writing – original draft, Writing – review & editing. Sofia Granados-Aparici: Investigation, Methodology, Writing – review & editing. Amparo López-Carrasco: Investigation, Methodology, Writing – review & editing. Miguel F. Segura: Methodology, Writing – review & editing. Manuel Salmerón-Sánchez: Methodology, Writing – review & editing. Inmaculada Noguera: Methodology, Writing – review & editing. Samuel Navarro: Investigation, Methodology, Validation, Writing – review & editing. Rosa Noguera: Conceptualization, Funding acquisition, Investigation, Methodology, Validation, Writing – original draft, Writing – review & editing. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Rosa Noguera reports financial support was provided by University of Valencia (UV), Foundation for Research of the Clinical Hospital of the Valencian Community (INCLIVA) and CIBERONC. Rosa Noguera reports a relationship with Foundation for Research of the Clinical Hospital of the Valencian Community that includes: non-financial support. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.