Unmasking phenotypic heterogeneity and extracellular matrix architecture in neuroblastoma through orthotopic xenograft modeling Authors: Meryem El Ouardi, Isaac Vieco-Martí, Sofía Granados-Aparici, Inmaculada Noguera, Samuel Navarro, Rosa Noguera Published: 2026 ABSTRACT Highlights•Original patient tumors display pronounced intratumoral heterogeneity of PHOX2B expression.•Advanced orthotopic neuroblastoma models recapitulate patient-like PHOX2B heterogeneity.•GTA-silk fibroin (GTA-sf) 3D hydrogel platform preserves tumor architecture in vivo.•Extracellular matrix stiffness and fibrillar organization are highly model-dependent.•Digital pathology enables quantitative, spatially resolved analysis of tumor-microenvironment interactions. Introduction Neuroblastoma (NB) is the most common extracranial solid tumor of childhood and accounts for approximately 15% of pediatric cancer-related deaths. It is an embryonal malignancy derived from neuroectodermal cells of the sympathoadrenal lineage, typically arising in the adrenal medulla, sympathetic ganglia, or paraganglia [1]. Despite advances in multimodal therapy, patients with high-risk NB (HR-NB) continue to face poor prognoses, with survival rates below 50% [2]. Tumor heterogeneity, cell lineage plasticity, and dynamic interactions with the tumor microenvironment (TME) play pivotal roles in driving therapy resistance and disease relapse in NB [3,4]. The diverse molecular and cellular landscape within NB tumors, arising from both intrinsic genetic variation and extrinsic TME influences, underscores the importance of understanding the crosstalk between tumor cells and their surrounding stroma [5,6]. Within this context, the extracellular matrix (ECM) emerges as a central regulator of tumor biology, influencing proliferation, differentiation, invasion, and metastatic dissemination [7,8]. ECM architecture shapes tumor structure and aggressiveness, with specific components, particularly the glycoprotein vitronectin (VN) and reticulin fibers, playing key roles in defining the structural and functional properties of the NB TME [[9], [10], [11]]. VN produced by malignant cells may contribute to the formation of mechanically stiff extracellular pathways, while three-dimensional (3D) culture systems, including hydrogels composed of gelatin methacryloyl (GelMA) and alginate methacrylate (AlgMA), allow tuning of matrix stiffness while preserving biocompatibility, facilitating studies of ECM-dependent tumor behaviors [12]. VN-based hydrogels can also recapitulate the biomechanical and cellular features of native NB tissue [13], and these biomechanical properties, including stiffness, may modulate oncogenic signaling and drive clonal selection, particularly in MYCN-amplified tumors [14]. Complementary in vivo approaches, such as orthotopic implantation of NB cells in immunodeficient mice, are essential to study tumor growth dynamics and TME remodeling under physiologically relevant conditions [15]. To bridge the gap between in vitro cell behavior and in vivo outcomes, we employed a sequential experimental pipeline integrating 2D cultures, non-VN-added 3D hydrogels, and orthotopic VN-knockout in vivo models to investigate how ECM cues influence NB cell identity. While 2D systems enable controlled mechanistic studies, they cannot reproduce the spatial and biomechanical complexity of native ECM-cell interactions. In contrast, 3D hydrogel-based models provide a tunable and architecturally relevant microenvironment that promotes sustained cell-ECM engagement [16]. During extended culture in these hydrogels, tumor cells actively remodel the matrix and undergo cytoskeletal and microenvironmental adaptations, which could represent a form of mechanostructural memory [17]. Although it remains to be determined whether such ECM-driven states are retained during orthotopic growth, 3D preconditioning provides a physiologically relevant context that could influence how cells subsequently integrate into and respond to the native TME, and may contribute to the phenotypic heterogeneity previously observed in NB cell lines and orthotopic xenograft models, providing a platform to dissect ECM-driven cellular plasticity [18,19]. Elucidation of these ECM-dependent cellular states provides a framework to explore how heterogeneity and lineage plasticity manifest in NB, in this case using paired-like homeobox 2b (PHOX2B) as a canonical marker of sympathoadrenal differentiation [20,21]. NB exhibits striking cellular heterogeneity and lineage plasticity, features that drive therapy resistance and disease progression [22,23]. This heterogeneity is currently understood within a framework comprising two major, interconvertible transcriptional states: a noradrenergic/adrenergic (NOR/ADRN) state and a mesenchymal (MES) state, each maintained by distinct core regulatory circuitries. PHOX2B is a component of the NOR regulatory program and a highly sensitive marker of neuroblastic lineage, although its expression alone does not define the complete NOR/ADRN-MES spectrum [24,25]. Increasing evidence indicates that heterogeneity arises from clonal diversification, with distinct subpopulations following divergent differentiation trajectories. Several studies have shown that NB subclones can adopt distinct lineage programs and coexist within the same tumor mass, reflecting dynamic transitions between NOR and MES identities governed by core transcriptional circuitries [26,27], and that disseminated cells at relapse often originate from minor subclones present at diagnosis [28]. Recent model systems have further revealed that NB cells can transition between differentiation states, underscoring the plastic and mosaic nature of tumor cell identity [29]. Given that cell fate and organization are strongly influenced by the ECM, we further examined the way PHOX2B heterogeneity aligns with ECM architecture. To systematically investigate these processes, we employed morphometric digital analysis to characterize cell phenotypes and behavior in both 2D and 3D scaffold-based NB models. Replicate cultures were analyzed to profile the expression patterns of PHOX2B, and to explore correlations with lineage markers and ECM features. This approach allows precise, quantitative assessment of cellular and microenvironmental parameters, providing proof of concept for the generation of clinically relevant in vitro models and enabling preclinical evaluation of therapeutic strategies. Materials and methods Gene expression and survival analysis Gene expression and survival analyses were performed using the R2 Genomics Analysis and Visualization Platform, developed by Jan Koster [30]. Analyses were conducted from the publicly available dataset “Tumor Neuroblastoma - SEQC-498-RPM-seqcnb1” (ps_avgpres_gse62564geo498_seqcnb1), which contains 498 NB samples profiled using both RNA-Seq and microarray platforms, available via GEO accession GSE62564 [31]. Expression intensities were normalized to raw values, with log2 transformation performed after adding a constant of 1 to avoid negative values. Expression levels of PHOX2B, MYCN, SOX9, vimentin (VIM), and tyrosine hydroxylase (TH) were evaluated across this cohort. Patients were classified into high and low expression groups based on optimal cutoff values automatically determined by the R2 platform. Kaplan-Meier survival analyses were carried out to assess the prognostic impact of each gene through overall survival (OS) and event-free survival (EFS), and between-group statistical significance was determined using the log-rank test. Clinical characteristics including age at diagnosis, disease stage, and risk group were obtained from the same dataset. Histopathological evaluation Histopathological analysis was performed on formalin-fixed, paraffin-embedded (FFPE) tumor specimens obtained from 16 independent NB patients. Tissue sections were stained with H&E to evaluate overall architecture and cellular morphology. Tumors were assessed for the characteristic small round blue cell morphology, including hyperchromatic nuclei, scant cytoplasm, and variable degrees of mitotic-karyorrhectic activity (MKI) consistent with poorly differentiated NB. Stroma was evaluated and classified as stroma-poor, and the degree of cellular differentiation was documented according to standard criteria. Diagnostic confirmation was supported by immunohistochemical (IHC) analysis of neuroblastic markers, including PHOX2B, synaptophysin, CD56, and TH. PHOX2B nuclear staining served as the principal diagnostic marker and semiquantitative assessment was performed to distinguish tumors with homogeneous expression (≥80% of tumor nuclei positive, n = 9) from those with heterogeneous expression (<80%, n = 7). Among the heterogeneous tumors, two were MYCN-amplified and five were MYCN-non-amplified. Based on these criteria, seven representative heterogeneous cases were selected for detailed histopathological morphometric characterization and comparative molecular analyses according to tissue availability and suitability for serial histological and immunohistochemical studies. The proportion of PHOX2B-positive cells was estimated across representative high-power fields, and expression patterns were recorded for each case. Tumor regions were classified according to the proportion of PHOX2B-positive nuclei. PHOX2B-negative foci were operationally defined as tumor regions of 0.2 mm² containing fewer than 35% PHOX2B-positive cells. This criterion was subsequently applied throughout the study for spatial analysis of PHOX2B-negative regions. Cell culture conditions Human NB cell lines SK-N-BE(2) (MYCN-amplified) and SH-SY5Y (ALK F1174L-mutated), representing 64% of high-risk NB genotypes [32,33], were obtained from ATCC (Supplementary Table S1). Cells were cultured in Iscove’s Modified Dulbecco’s Medium (IMDM) with 10% Fetal Bovine Serum (FBS), 1% Insulin-Transferrin-Selenium Growth Supplement (ITS-G), 1% penicillin/streptomycin, and 0.2% Plasmocin™ at 37 °C, 5% CO₂. Medium was replaced every two days, and cells were passaged at ∼90% confluence before 3D embedding. PDX1 and PDX2 are two independent MYCN-amplified patient-derived orthotopic xenograft (PDX) models established from high-risk NB tumors obtained from patients treated at the Hospital Clínico Universitario de Valencia. Tumor specimens were processed and diagnosed by one of the co-authors (SN) at the Department of Pathology of the Hospital Clínico Universitario de Valencia. Tumor material was subsequently transferred to Lund University (Sweden), where the orthotopic PDX models were established by Braekeveldt, Bexell and colleagues, as previously described [[34], [35], [36]] (Supplementary Table S2). PDX-derived cells were maintained in Dulbecco’s Modified Eagle Medium DMEM-GlutaMAX™/F12 (1:1) supplemented with 2% B27 (-vitamin A), 1% penicillin/streptomycin, 40 ng/mL basic fibroblast growth factor (bFGF), and 20 ng/mL EGF under the same conditions. Primary NB samples were obtained from patient biopsies after informed consent was recieved. The procedures for processing and short-term culture of NB explants were based on previously established methods for patient-derived NB models [35,36]. All reagents used for the cell culture procedures are detailed in Supplementary Table S3. Composite hydrogel crafting and characterization Tyramine-modified gelatin (GTA) was synthesized following a previously described method by EDC/NHS coupling of gelatin with tyramine, followed by dialysis and lyophilization [37]. GTA (1%w/v) was UV-sterilized and dissolved in serum-free medium at 50 °C, then mixed with silk fibroin (sf, 3% w/v). After trypsinization and counting, approximately 120,000 cells (240,000 cells for PDX) were suspended in the GTA-sf solution. Gelation was induced using horseradish peroxidase (20 U/mL) and hydrogen peroxide (H₂O₂; final concentration 0.01%, non-cytotoxic; 2 μL) in 24-well plates at 37 °C for 1 h. Hydrogels were cultured for 3 weeks at 37 °C and 5% CO₂, with medium changes every 2–3 days. Mechanical properties were assessed by uniaxial compression testing using a Zwick/Roell Z0.5 mechanical tester (Zwick-Roell, Germany). Young’s modulus was calculated via linear regression within the 5–15% strain interval in the linear elastic region [38]. All reagents used for hydrogel fabrication are detailed in Supplementary Table S4. In vivo models Orthotopic NB models were established as previously reported [9]. For orthotopic inoculation, RAG1⁻/⁻VN⁻/⁻ mice (RAG1⁻/⁻: B6.129S7-Rag1tm1Mom/J; VN⁻/⁻: B6.129S2(D2)-Vtntm1Dgi/J), 4–6 weeks of age, of both sexes, were anesthetized with isoflurane (4–5% for induction and 1–2% for maintenance) The number of mice orthotopically inoculated for each NB model is detailed in Supplementary Table S5. Tumor cells were implanted into the left adrenal gland via a small flank incision. For the 3D condition, cells were cultured for 3 weeks within GTA-sf hydrogels and subsequently recovered from the hydrogels prior to inoculation. Approximately 1 × 10⁶ recovered cells were then suspended in Matrigel/dPBS and inoculated using the same procedure as for the 2D condition. Therefore, the only experimental difference between groups was the culture platform before implantation (conventional 2D culture versus 3D hydrogel preconditioning). Surgical sites were sutured, and mice received postoperative analgesia (buprenorphine, 0.1 mg/kg, subcutaneously). All neuroblastoma cell lines and PDX models used for orthotopic implantation were previously transduced to stably express firefly luciferase, enabling non-invasive longitudinal monitoring of tumor growth by bioluminescence imaging (BLI). Tumor progression was first monitored by BLI using an in vivo imaging system (IVIS Spectrum, PerkinElmer) at day 10 post inoculation. Upon positive tumor detection, BLI monitoring was subsequently performed every 10 days to evaluate tumor growth dynamics. Once tumors reached endpoint criteria, mice were euthanized by isoflurane overdose followed by cervical dislocation. Tumors, together with the contralateral kidney and adrenal gland, liver, lungs, and spleen, were collected and fixed in 10% neutral-buffered formalin for FFPE. Fig. 1 shows a schematic overview of the experimental workflow, from cell culture and hydrogel fabrication to orthotopic inoculation, longitudinal BLI monitoring, and histological and digital image analyses. All reagents, materials, and equipment used for the in vivo procedures are detailed in Supplementary Table S6.Fig. 1Schematic overview of the xenograft experimental workflow and tumor growth monitoring. Neuroblastoma (NB) cell lines (SK-N-BE(2), SH-SY5Y) cultured as two-dimensional (2D) monolayers or three-dimensional (3D) hydrogel models, along with patient-derived xenograft 3D hydrogels (PDX1, PDX2), were engrafted into immunodeficient RAG1⁻/⁻; VN⁻/⁻ mice. Tumor growth was monitored longitudinally by in vivo bioluminescence imaging (BLI) until detectable signals emerged, followed by tissue collection for analysis. Histopathological characterization included H&E staining, IHC, and digital image analysis. Representative bioluminescence images illustrating tumor progression over time are shown for SK-N-BE(2), SH-SY5Y, PDX1, and PDX2 xenografts, together with quantification of tumor burden expressed as total flux (photons/second) across time. Schematic overview of the xenograft experimental workflow and tumor growth monitoring. Neuroblastoma (NB) cell lines (SK-N-BE(2), SH-SY5Y) cultured as two-dimensional (2D) monolayers or three-dimensional (3D) hydrogel models, along with patient-derived xenograft 3D hydrogels (PDX1, PDX2), were engrafted into immunodeficient RAG1⁻/⁻; VN⁻/⁻ mice. Tumor growth was monitored longitudinally by in vivo bioluminescence imaging (BLI) until detectable signals emerged, followed by tissue collection for analysis. Histopathological characterization included H&E staining, IHC, and digital image analysis. Representative bioluminescence images illustrating tumor progression over time are shown for SK-N-BE(2), SH-SY5Y, PDX1, and PDX2 xenografts, together with quantification of tumor burden expressed as total flux (photons/second) across time. Immunohistochemistry Xenograft tumors and murine organs were fixed in 10% neutral-buffered formalin for 24–48 h. Tissues were processed using standard histological procedures, including dehydration through a graded ethanol series, clearing in xylene, and infiltration and embedding in paraffin wax. Paraffin-embedded tissue blocks were sectioned at a thickness of 3 µm. Routine morphology was assessed on hematoxylin and eosin (H&E) stained sections. IHC staining was performed on representative tumor and organ sections using antibodies directed against VN, SOX9, PHOX2B, TH, VIM, and the endothelial marker ERG. Antigen retrieval was carried out in citrate buffer (pH 6.0) or Tris-EDTA buffer (pH 9.0) depending on the primary antibody. Detection was achieved with the EnVision+ System-HRP, using 3,3′-diaminobenzidine (DAB) as the chromogen, and slides were counterstained with hematoxylin. Negative controls included omission of the primary antibody or substitution with isotype-matched IgG, and known positive tissues served as external controls. Details of all primary antibodies, including source, clone, catalog number, and working dilution, are provided in Supporting Information. Digital analysis H&E-stained sections and the corresponding IHC stained slides were digitized using a Ventana iScan HT scanner at 20× magnification (0.465 μm/pixel). All whole-slide images (WSIs) were analyzed using QuPath (v0.5.1) open-source digital pathology software [39]. For each H&E WSI, color deconvolution was first performed using QuPath built-in stain separation tool. Tumor and non-tumor regions were manually annotated, while artefact areas were systematically excluded from analysis. Neuroblasts were segmented from H&E and IHC images using the StarDist extension [[40], [41], [42]], with the inference resolution standardized to 0.25 μm/pixel; images acquired at lower resolutions were rescaled accordingly. For IHC analyses, positive cells were detected using marker-specific optical density thresholds. Thresholds were kept constant within each staining batch and were validated by a pathologist to ensure accurate discrimination between positive and negative cells. Quantification was expressed as the percentage of positive cells relative to the total number of detected cells. For PHOX2B, negative foci were defined as tumor regions of approximately 0.2 mm² with a positivity rate below 35%. Interactive alignment of consecutive tissue sections was then performed to determine the percentage of positive cells within the same PHOX2B-negative foci for the plasticity markers VIM, SOX9, TH, and ERG. Finally, spatial relationships between NB cells were evaluated using the long-radius expansion method implemented in JTS, where clusters within or intersecting the expanded radius were considered neighbors. VN localization was further categorized based on staining intensity derived from the pixel classifier output: territorial VN (tVN), defined by strongly stained (high-intensity) VN-positive pixels, and interterritorial VN (interVN), characterized by low-to-medium intensity VN-positive pixels. Intensity cutoffs distinguishing these categories were fixed per staining batch and validated by a pathologist. Reticulin-stained sections were analyzed for fiber organization using the open-source tools CT-FIRE and CurveAlign. Representative regions of interest (ROIs; 1000 × 1000 µm) were exported from QuPath and processed in CT-FIRE for individual fiber extraction, followed by CurveAlign analysis to quantify fiber length, orientation, and alignment. Identical parameters were applied across all samples, and fiber segmentation quality was verified by visual inspection. Quantitative outputs were used for downstream statistical analyses. Finally, cell-to-cell spatial relationships were evaluated using the long-radius expansion method implemented in DANEELpath [43], based on the JTS Topology Suite, whereby cells located within or intersecting the expanded radius of a given cell were considered neighbors. Statistical analyses Statistical analyses were performed using R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). Data normality was assessed using the Shapiro-Wilk test, and homogeneity of variances was evaluated using Levene's test. When the assumptions for parametric testing were satisfied, between-group comparisons were performed using one-way or two-way analysis of variance (ANOVA), followed by the appropriate post hoc multiple-comparison test. Paired two-tailed Student’s t-tests were used for paired comparisons of PHOX2B-positive and PHOX2B-negative regions. Statistical significance was defined as p < 0.05. Results Context-dependent associations between lineage marker expression and clinical outcome We analyzed gene expression profiles associated with lineage differentiation and plasticity in NB using publicly available patient datasets, including the SEQC (GSE62564) cohorts. Consistent with previous studies highlighting NB cellular heterogeneity and lineage plasticity [22,23], we focused on PHOX2B, a key marker of sympathoadrenal differentiation [20,21]. Kaplan-Meier analyses suggested context-dependent trends in the relationship between PHOX2B expression and patient outcome (Fig. 2A), although these associations did not remain statistically significant after Bonferroni correction for multiple comparisons. In HR-NB cases, higher PHOX2B expression showed a trend toward better OS, consistent with its association with more differentiated, less plastic cell states [44]. Conversely, in MYCN-amplified tumors, higher PHOX2B expression showed a tendency toward poorer outcome, potentially reflecting the expansion of proliferative PHOX2B-positive progenitors [45]. Similarly, higher SOX9 and VIM expression showed trends toward poorer survival, whereas TH expression showed a weaker association with patient outcome [[46], [47], [48], [49]]. Although these observations did not reach statistical significance after correction for multiple comparisons, they are consistent with the potential relevance of the balance between neuroendocrine and mesenchymal transcriptional programs, rather than the abundance of a single lineage marker, in clinical outcome [[37], [38], [39], [40]]. Given that ALK overexpression and MYCN amplification frequently co-occur, and their interaction influences tumor behavior, we further assessed their relationship with lineage markers. In MYCN-amplified HR-NB tumors from the SEQC cohort, ALK expression showed a significant positive correlation with SOX9 levels (Fig. 2B). Tumors with ALK mutations but without MYCN amplification displayed less aggressive features. SH-SY5Y cells, which harbor an activating ALK mutation, represented a relevant model for studying ALK-dependent behavior, including regulation of SOX9 and its interaction with ECM components. Together, these findings indicate that co-analysis of oncogenic drivers and lineage markers provides mechanistic insight into the relationship of NB identity and plasticity with ECM interactions and tumor aggressiveness.Fig. 2Kaplan-Meier survival and gene expression analysis of neuroblastoma (NB) patients. (A) Kaplan-Meier overall survival (OS) curves for NB patients stratified by gene expression and clinical risk. Top row: Survival analysis stratified by HR-NB patients based on the expression of PHOX2B, tyrosine hydroxylase (TH), SOX9 and vimentin (VIM). Bottom row: Kaplan-Meier curves according to MYCN-amplified NB patients for the same genes, showing OS differences between high and low expression groups. Log-rank p-values are indicated on each plot. (B) Scatter plot showing a positive moderate correlation between SOX9 and ALK expression (R = 0.351, p = 5.95×10⁻⁴) in MYCN-amplified NB samples (n = 92). Samples are ordered by SOX9 expression (log₂-transformed). Expression data was retrieved from the SEQC neuroblastoma dataset (GEO accession GSE62564; Tumor Neuroblastoma - SEQC - 498 - RPM - seqcnb1)]. Kaplan-Meier survival and gene expression analysis of neuroblastoma (NB) patients. (A) Kaplan-Meier overall survival (OS) curves for NB patients stratified by gene expression and clinical risk. Top row: Survival analysis stratified by HR-NB patients based on the expression of PHOX2B, tyrosine hydroxylase (TH), SOX9 and vimentin (VIM). Bottom row: Kaplan-Meier curves according to MYCN-amplified NB patients for the same genes, showing OS differences between high and low expression groups. Log-rank p-values are indicated on each plot. (B) Scatter plot showing a positive moderate correlation between SOX9 and ALK expression (R = 0.351, p = 5.95×10⁻⁴) in MYCN-amplified NB samples (n = 92). Samples are ordered by SOX9 expression (log₂-transformed). Expression data was retrieved from the SEQC neuroblastoma dataset (GEO accession GSE62564; Tumor Neuroblastoma - SEQC - 498 - RPM - seqcnb1)]. PHOX2B intratumoral heterogeneity Examination of primary patient biopsies of poorly differentiated NB revealed highly cellular tumors with microscopic areas of neuropil, corresponding to regions of sparse neuritic processes. A total of 16 NB cases were analyzed, of which seven displayed a mosaic pattern of PHOX2B expression, characterized by intermingled PHOX2B-positive and PHOX2B-negative tumor cells (Supplementary Table S7). In some of these cases, we could identify well-defined focal regions of PHOX2B negativity, measuring 0.2 mm². Overall, PHOX2B-positive (mosaic-like) regions contained a mean of 71.5% PHOX2B-positive cells, whereas PHOX2B-negative foci contained a mean of 28.5% PHOX2B-positive cells. Based on this distribution, tumor regions of 0.2 mm² containing fewer than 35% PHOX2B-positive cells were operationally defined as PHOX2B-negative foci and selected for all subsequent orthotopic analyses (Supplementary Figure S1). To exclude the possibility that reduced PHOX2B expression in these regions was primarily associated with apoptotic or degenerative changes, CASP3 expression was evaluated in corresponding tissue areas. CASP3 positivity remained low within PHOX2B-negative tumor regions, indicating that the reduced PHOX2B expression observed in these foci was not attributable to increased apoptotic cell death (Supplementary Figure S2). Morphometric analysis demonstrated significant differences between tumor subpopulations in nuclear size and cell shape, with PHOX2B-negative cells exhibiting smaller nuclear areas and slightly higher circularity than PHOX2B-positive cells. PHOX2B expression was also examined in the four NB models after 3 weeks of culture in GTA-sf hydrogels (Supplementary Table S7). PHOX2B-positive cells were detected in only a minority of tumor clusters. Among these PHOX2B-positive clusters, the staining pattern was heterogeneous: some showed diffuse PHOX2B expression involving most tumor cells, whereas others displayed focal expression restricted to a subset of cells within the cluster (Supplementary Figure S3). We next examined whether the corresponding orthotopic xenografts recapitulated the PHOX2B heterogeneity observed in patient tumors. Xenografts derived from 2D cell suspensions showed relatively uniform PHOX2B staining, failing to reproduce patient-like heterogeneity. Conversely, 3D hydrogel-derived xenografts displayed interspersed PHOX2B-positive and PHOX2B-negative zones, closely mimicking the mosaic architecture of patient tumors (Fig. 3A); however, PDX2 samples are not shown, as they lacked a mosaic PHOX2B expression pattern and remained largely homogeneous (Supplementary Figure S4). These results indicate that only specific 3D preclinical models recapitulate the cellular diversity of PHOX2B expression observed in patient tumors. Nuclear morphometric analysis further revealed model-dependent differences between PHOX2B-positive and PHOX2B-negative cells. In SH-SY5Y and PDX1 tumors, negative cells exhibited smaller nuclear areas, whereas no size differences were detected in SK-N-BE(2) (Fig. 3B). Nuclear circularity remained unchanged in PDX1 but was increased in SK-N-BE(2) and SH-SY5Y negative cells (Fig. 3C). In conjunction, these findings suggest that PHOX2B status correlates with distinct nuclear morphologies in a model-specific manner, emphasizing the phenotypic complexity captured by 3D-derived xenografts.Fig. 3Model-dependent PHOX2B spatial heterogeneity. (A) Spatial distribution of PHOX2B-negative and -positive NB cells across orthotopic tumors of SH-SY5Y, SK-N-BE(2), either in 3D Hydrogel or 2D model, PDX1 and original tumors (OT). (B, C) Distribution of nuclear area and circularity in PHOX2B-negative (■) and -positive (■) cells across all NB cell lines. (D) Representative regions showing PHOX2B-negative foci (■; 35% positivity) and PHOX2B-positive areas (■; 75% positivity) for each model. Scale bars: 50 µm. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using a paired two-tailed Student’s t-test; **p < 0.01, ***p < 0.001; ns, not significant. Model-dependent PHOX2B spatial heterogeneity. (A) Spatial distribution of PHOX2B-negative and -positive NB cells across orthotopic tumors of SH-SY5Y, SK-N-BE(2), either in 3D Hydrogel or 2D model, PDX1 and original tumors (OT). (B, C) Distribution of nuclear area and circularity in PHOX2B-negative (■) and -positive (■) cells across all NB cell lines. (D) Representative regions showing PHOX2B-negative foci (■; 35% positivity) and PHOX2B-positive areas (■; 75% positivity) for each model. Scale bars: 50 µm. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using a paired two-tailed Student’s t-test; **p < 0.01, ***p < 0.001; ns, not significant. Plasticity and structured heterogeneity In SH-SY5Y tumors, PHOX2B-negative foci showed substantially reduced expression of the noradrenergic marker TH together with increased SOX9 and VIM expression, consistent with attenuation of the noradrenergic program and acquisition of mesenchymal-like features (Fig. 4B). This pattern suggests that PHOX2B downregulation in SH-SY5Y cells is associated with lineage plasticity and acquisition of non-adrenergic features. No upregulation of mesenchymal markers was found in PHOX2B-negative regions in SK-N-BE(2) and PDX1 tumors, however, which instead retained TH expression and adrenergic identity despite reduced PHOX2B staining (Figs. 4A and 4C). These findings indicate that spatial PHOX2B heterogeneity encompasses distinct biological contexts: in SK-N-BE(2) and PDX1, PHOX2B-negative areas likely represent lineage-stable states, whereas in SH-SY5Y, PHOX2B loss was associated with a marker profile compatible with mesenchymal-like plasticity in SH-SY5Y.Fig. 4Cell-state marker heterogeneity between PHOX2B-positive and -negative foci. Representative IHC staining of PHOX2B, tyrosine hydroxylase (TH), SOX9, and vimentin (VIM) in PHOX2B-negative and PHOX2B-positive regions of SK-N-BE(2) (A), SH-SY5Y (B), and PDX1 (C) NB xenografts. Insets show magnified areas highlighting marker distribution within individual cells. Bottom panels show quantification of marker positivity (% of positive cells) for each tumor model in negative foci versus positive area. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using a paired two-tailed Student’s t-test; *p < 0.05, **p < 0.01; ns, not significant. Cell-state marker heterogeneity between PHOX2B-positive and -negative foci. Representative IHC staining of PHOX2B, tyrosine hydroxylase (TH), SOX9, and vimentin (VIM) in PHOX2B-negative and PHOX2B-positive regions of SK-N-BE(2) (A), SH-SY5Y (B), and PDX1 (C) NB xenografts. Insets show magnified areas highlighting marker distribution within individual cells. Bottom panels show quantification of marker positivity (% of positive cells) for each tumor model in negative foci versus positive area. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using a paired two-tailed Student’s t-test; *p < 0.05, **p < 0.01; ns, not significant. Having established that PHOX2B-negative regions in SH-SY5Y xenografts are enriched for plasticity markers such as SOX9 and VIM, we next sought to determine whether these dedifferentiated tumor niches also differed in vascularization. To this end, we analyzed ERG, an endothelial marker, across all tumor models (Fig. 5). ERG-positive cell density was higher in PHOX2B-negative foci compared with PHOX2B-positive regions in all models. Interestingly, while SH-SY5Y PHOX2B-negative regions exhibited strong dedifferentiation, reflected by elevated SOX9 and VIM, SK-N-BE(2) and PDX tumors showed less pronounced plasticity despite similar ERG enrichment. This suggests that increased vascularization of PHOX2B-negative niches may occur independently of the degree of dedifferentiation. Moreover, higher ERG-positive cell density implies that these regions are likely better perfused and less hypoxic, potentially supporting elevated local metabolic and proliferative activity.Fig. 5Correlation between PHOX2B-negativity and vascularization. Representative density maps of PHOX2B-negative cells (■) and ERG-positive cells (■) in SK-N-BE(2) (A), SH-SY5Y (B), and PDX1 (C). Scale bars: 150 µm. Inset: Magnified images of PHOX2B and ERG expression in both PHOX2B-negative foci (■) and PHOX2B-positive areas (■). Scale bars: 15 µm. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using paired two-tailed Student’s t-test; **p < 0.01, ***p < 0.001. Correlation between PHOX2B-negativity and vascularization. Representative density maps of PHOX2B-negative cells (■) and ERG-positive cells (■) in SK-N-BE(2) (A), SH-SY5Y (B), and PDX1 (C). Scale bars: 150 µm. Inset: Magnified images of PHOX2B and ERG expression in both PHOX2B-negative foci (■) and PHOX2B-positive areas (■). Scale bars: 15 µm. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using paired two-tailed Student’s t-test; **p < 0.01, ***p < 0.001. Hydrogel stiffness and tumor growth Mechanical characterization of the hydrogels prior to grafting showed that SH-SY5Y and PDX2 hydrogels were the stiffest (1.7–2.2 kp), while SK-N-BE(2) and PDX1 hydrogels were softer (1–1.4 kp) (Fig. 6A). Upon inoculation, softer hydrogels supported the largest xenograft areas, consistent with a more permissive microenvironment, whereas stiffer hydrogels limited tumor volume. Morphometric analysis supported these observations: SK-N-BE(2) tumors had low cell density but high cell circularity, indicating dispersed yet regularly shaped cells (Fig. 6B and 6C). SH-SY5Y xenografts also combined low nearby cell count with high circularity, reflecting moderately cohesive and round cells (Fig. 6B and 6C). In contrast, PDX2 tumors exhibited high neighbor density but lower circularity, corresponding to compact but irregularly shaped cells, whereas PDX1 xenografts displayed the highest neighbor density and lowest circularity, suggesting densely packed cells (Fig. 6B and 6C). Quantitative analysis of tumor xenografts further revealed marked differences in tissue composition across models. PDX2 xenografts were highly enriched in tumor cell niches, with minimal stromal content and necrosis, a pattern similarly observed in SK-N-BE(2) tumors. Conversely, SH-SY5Y xenografts contained a larger stromal fraction, while PDX1 tumors exhibited reduced tumor content and increased necrotic areas (Fig. 6D). To gain further insight into the extracellular environment underlying these structural differences, we quantified VN expression. VN staining was highest in SH-SY5Y xenografts, within both tVN and interVN regions, whereas PDX1 tumors exhibited the lowest VN signal (Fig. 6E and 6F).Fig. 6Model-dependent extracellular matrix architecture. SK-N-BE(2), SH-SY5Y, PDX1, and PDX2 xenografts. (A) Correlation between tumor area and 3D hydrogel stiffness. (B) Box plot comparing cell circularity across the NB models. (C) Quantification of neighboring NB cells. (D) Proportion of tumor cells, stromal cells, and necrotic cells. (E) Territorial (tVN) and interterritorial (interVN) VN expression, depicted as the percentage of stained area. (F) Representative VN immunostaining images showing VN expression patterns for each model. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using one-way ANOVA followed by Tukey’s multiple-comparisons test; **p < 0.01, ***p < 0.001. Model-dependent extracellular matrix architecture. SK-N-BE(2), SH-SY5Y, PDX1, and PDX2 xenografts. (A) Correlation between tumor area and 3D hydrogel stiffness. (B) Box plot comparing cell circularity across the NB models. (C) Quantification of neighboring NB cells. (D) Proportion of tumor cells, stromal cells, and necrotic cells. (E) Territorial (tVN) and interterritorial (interVN) VN expression, depicted as the percentage of stained area. (F) Representative VN immunostaining images showing VN expression patterns for each model. For each experimental condition, two predefined tumor regions were analyzed in four independent orthotopic tumors (one tumor per mouse). Data are presented as mean ± SEM. Statistical significance was assessed using one-way ANOVA followed by Tukey’s multiple-comparisons test; **p < 0.01, ***p < 0.001. Consistent with the distinct growth patterns of the models, quantitative analysis of reticulin fibers revealed marked differences in ECM organization (Fig. 7). SH-SY5Y xenografts exhibited the longest fibers with the highest straightness and a narrow angular distribution, indicative of preferential fiber alignment (Fig. 7B and 7E). In contrast, SK-N-BE(2) and PDX1 tumors were characterized by significantly shorter fibers with reduced straightness and broader angle distributions, reflecting a more curved and disorganized fibrillary network (Fig. 7A, 7C, and 7E) consistent with a mechanically heterogeneous ECM. Fiber width also differed significantly across models: PDX2 xenografts displayed the thickest fibers, whereas SH-SY5Y tumors contained the thinnest fibers, with SK-N-BE(2) and PDX1 exhibiting intermediate widths (Fig. 7E). Despite their high stiffness, PDX2 tumors combined increased fiber width with relatively high straightness, intermediate fiber length, and a narrow angular dispersion, consistent with a compact and directionally organized ECM associated with dense tumor cell niches and limited stromal content (Fig. 7D and 7E). Overall, the distributions of fiber length, straightness, orientation, and width demonstrate that ECM stiffness and fibrillary architecture are not uniform across NB models but instead reflect tumor-intrinsic programs of microenvironmental remodeling. Collectively, these findings indicate that the interplay between ECM organization, stromal balance, and tumor cell niche density critically shapes the architectural and pathological features of NB in vivo.Fig. 7Analysis of reticulin expression pattern using CT FIRE and CurveAlign software. (i) Selected regions from reticulin immunostaining within orthotopic NB tumors of SK-N-BE(2) (A), SH-SY5Y (B), PDX1 (C), and PDX2 (D) (ROIs: 1000×1000 µm). (ii) Reticulin fiber masks generated from raw ROI images. (iii) Graphical output from CT FIRE showing automatic extraction of reticulin fibers. (iv) Angle compass plots displaying the orientation of fibers. (E) Graphs are showing straightness, length, angle and width of reticulin fibers calculated using CurveAlign. Analysis was performed on five randomly selected regions of interest per tumor from four independent orthotopic xenografts for each experimental model (n = 4 tumors/model; 20 ROIs/model). Data are presented as mean ± SEM. Statistical significance was assessed using one-way ANOVA followed by Tukey’s multiple-comparisons test; **p < 0.01, ***p < 0.001; ns, not significant. Analysis of reticulin expression pattern using CT FIRE and CurveAlign software. (i) Selected regions from reticulin immunostaining within orthotopic NB tumors of SK-N-BE(2) (A), SH-SY5Y (B), PDX1 (C), and PDX2 (D) (ROIs: 1000×1000 µm). (ii) Reticulin fiber masks generated from raw ROI images. (iii) Graphical output from CT FIRE showing automatic extraction of reticulin fibers. (iv) Angle compass plots displaying the orientation of fibers. (E) Graphs are showing straightness, length, angle and width of reticulin fibers calculated using CurveAlign. Analysis was performed on five randomly selected regions of interest per tumor from four independent orthotopic xenografts for each experimental model (n = 4 tumors/model; 20 ROIs/model). Data are presented as mean ± SEM. Statistical significance was assessed using one-way ANOVA followed by Tukey’s multiple-comparisons test; **p < 0.01, ***p < 0.001; ns, not significant. Discussion In this study, we integrated 2D cultures and 3D GTA-silk fibroin hydrogels following orthotopic implantation to examine the way extracellular matrix (ECM) architecture and biomechanical cues influence lineage identity and phenotypic heterogeneity in NB. PHOX2B heterogeneity emerged as a consistent feature across patient samples and several models, but its biological meaning varied substantially depending on genetic background and microenvironmental context. The observation that unlike 2D-derived xenografts, 3D hydrogel-derived xenografts were able to reproduce the mosaic PHOX2B expression patterns seen in patient tumors underscores the importance of structurally complex matrices for maintaining NB cellular diversity. Notably, after 3 weeks of in vitro culture, PHOX2B expression was detected in only a minority of tumor clusters, with heterogeneous intracluster patterns ranging from positivity restricted to a subset of cells to more diffuse expression. This indicates that the mosaic PHOX2B architecture subsequently observed in selected 3D-derived xenografts was not fully established during in vitro culture. The duration of 3D culture should also be considered as a potential factor influencing PHOX2B expression, as lineage-associated phenotypes may evolve during prolonged exposure to the hydrogel microenvironment. Together, these observations suggest that the phenotypic consequences of 3D preconditioning may continue to evolve following implantation through interactions with the in vivo tumor microenvironment [50,51]. Long-established neuroblastoma cell lines may not fully recapitulate the morphology, physiology, and three-dimensional architecture of primary tumors, as prolonged in vitro culture and repeated passaging can promote phenotypic and molecular divergence from the tumor of origin. This represents a limitation of the present study. To partially address this constraint, we employed a multimodel experimental design combining established cell lines with patient-derived orthotopic xenografts (PDXs) and primary patient specimens. This complementary approach allowed observations from established cell lines to be interpreted alongside biologically more representative models and enabled the identification of both shared and model-dependent features. Nevertheless, findings derived from established cell lines should be interpreted within the context of these inherent limitations. Within this context, our findings are also compatible with the concept of ECM-driven mechanostructural memory, whereby prolonged exposure to defined biomechanical environments induces stable cellular states that persist even after transfer to a new microenvironment. The emergence of patient-like mosaic PHOX2B patterns in xenografts derived from 3D hydrogel-preconditioned cells, even though PHOX2B expression was restricted to a minority of tumor clusters during in vitro culture, raises the possibility that ECM-derived cues prime cellular states that are subsequently shaped or selected within the in vivo microenvironment to influence tumor organization and lineage heterogeneity. Such memory effects have been described in other solid tumors, where stiffness and matrix architecture promote long-lasting transcriptional and cytoskeletal adaptations that influence in vivo behavior. In this context, our data indicate that hydrogel preconditioning may encode biomechanical information which conditions subsequent tumor organization and lineage heterogeneity in NB [52]. Beyond the influence of 3D preconditioning, the biological significance of PHOX2B-negative foci differed substantially between models. SH-SY5Y xenografts displayed PHOX2B downregulation together with reduced TH expression and increased SOX9 and VIM expression, a marker profile compatible with attenuation of the noradrenergic program and acquisition of mesenchymal-like features. However, PHOX2B alone does not define the complete NOR/ADRN-MES spectrum, and SOX9 and VIM are not exclusive markers of MES neuroblastoma cells. Therefore, these findings should be interpreted as evidence of spatially organized lineage plasticity rather than as a definitive transcriptional classification of individual tumor cells. This marker profile is consistent with features previously associated with NOR-to-MES plasticity and bidirectional adrenergic-mesenchymal transitions described in NB, which have been implicated in therapy resistance and immune evasion [53,54], and further highlights the sensitivity of SH-SY5Y cells to ECM-dependent cues. These findings support the concept that mesenchymal-associated features in NB are heterogeneous and strongly influenced by the microenvironment. In this respect, our observations are also consistent with reports describing CD133-positive stem-like subpopulations as a rare and heterogeneous component of aggressive NB rather than a uniform tumor state [55]. In contrast, SK-N-BE(2) and PDX1 models exhibited PHOX2B-negative regions that retained adrenergic identity, indicating that reduced PHOX2B expression does not necessarily correspond to a lineage transition. Accordingly, PHOX2B heterogeneity should be interpreted as spatial heterogeneity in neuroblastic lineage-marker expression rather than as a direct surrogate of the NOR/ADRN–MES classification. Consistent with this interpretation, mesenchymal-like features were observed only in SH-SY5Y xenografts, whereas SK-N-BE(2) and PDX1 retained noradrenergic characteristics despite reduced PHOX2B expression. Together, these observations indicate that spatial histology complements transcriptional cell-state classifications by identifying where lineage-associated phenotypes arise within the tissue architecture and how they relate to vascular and extracellular-matrix niches. Vascular niches have been reported to support stem-like or mesenchymal-like populations in NB and other cancers [56,57]. In our models, PHOX2B-negative niches showed increased ERG positivity, with ERG-positive nuclei occurring in different spatial contexts. While many were associated with recognizable vascular structures, isolated ERG-positive cells and small ERG-positive cell clusters were also observed within PHOX2B-negative foci. The identity of these cells cannot be definitively established from ERG staining alone; this would require additional endothelial markers or multiplex analyses. Thus, although their distribution is consistent with an increased vascular or perivascular component, the possibility that a subset represents tumor cells with endothelial-like features cannot be excluded. Consequently, PHOX2B-negative regions likely comprise variable proportions of tumor cells with altered lineage-marker expression together with non-neoplastic stromal and vascular/perivascular components. Interestingly, while SH-SY5Y PHOX2B-negative regions exhibited a marker profile compatible with mesenchymal-like plasticity, SK-N-BE(2) and PDX1 showed a similar increase in ERG positivity without concomitant upregulation of SOX9 or VIM. These findings suggest that ERG enrichment and lineage plasticity do not necessarily represent the same biological phenomenon and highlight the value of spatial histology for identifying distinct cellular populations within PHOX2B-negative niches that warrant further phenotypic characterization. Beyond the cellular composition of these niches, the mechanical and structural properties of the ECM also emerged as important determinants of tumor architecture. Softer hydrogels produced larger xenografts, whereas stiffer matrices constrained tumor expansion, in line with previous reports showing that ECM stiffness modulates tumor growth and cellular organization [58,59]. Reticulin fiber quantification showed that SH-SY5Y tumors developed long, aligned fibers consistent with a mechanically structured microenvironment, whereas SK-N-BE(2) and PDX1 tumors contained shorter and more curved fibers associated with disorganized ECM remodeling, echoing findings that fiber alignment influences cell clustering and intratumoral heterogeneity [60,61]. PDX2 xenografts formed compact tumor cell clusters within highly aligned, thickened fibers, indicating a distinct pattern of stromal organization. Altogether, these results demonstrate that ECM stiffness, fiber organization, and cellular composition jointly regulate NB structure and phenotypic diversity, extending prior evidence from other solid tumors and 3D scaffold models to NB. Conclusion This study adds to the increasing body of evidence that advanced 3D tumor models are essential for narrowing the gap between simplified in vitro systems and the complex biological behavior observed in NB patients. Consistent with recent reviews on 3D models, our results demonstrate that preconditioning NB cells within ECM-stiff hydrogels induces a form of mechanostructural memory that preserves structural and phenotypic features lost in conventional 2D cultures. Notably, the relative size (5–6 mm2) and optimum thickness (1 mm) of our hydrogel scaffolds enable the formation of reproducible multicellular clusters which better recapitulate the organizational complexity of NB tumors. Introducing cells into the in vivo environment only after adaptation to relevant ECM-related and mechanical cues enhances the ability of orthotopic xenografts to reproduce patient-like tumor organization, lineage heterogeneity, and microenvironmental interactions. By more faithfully reflecting the native TME, this approach improves the translational relevance of in vivo models for studies on therapeutic resistance and prognostic evaluation, while its increased reproducibility and lower experimental numbers help reduce animal use. Overall, integrating physiologically informed, large-scale hydrogel scaffolds with optimized orthotopic models represents a significant step toward generating more accurate and ethically responsible NB models that facilitate the development of personalized therapeutic strategies. Funding This research was funded by the Instituto de Salud Carlos III (ISCIII) and the European Regional Development Fund (FEDER), grant number PI20/01,107; the Fundación Neuroblastoma (FNB), grant number PRV/00,166; and CIBERONC, contract CB16/12/00,484. I.V.-M. was supported by a predoctoral fellowship from the Ministry of Science, Innovation and Universities of Spain (FPU20/05,344). Institutional review board statement This study was approved by the Research Ethics Committee of the Clinic Hospital of Valencia (reference 2020/025, Act: 372, 30 September 2021). The animal experiments were carried out in accordance with the standards and care approved by the institutional ethical animal care committee (reference 2021/VSC/PEA/0052). Informed consent statement Participants or their family members/legal guardians provided written informed consent for histological studies performed in our laboratory (reference B.0000,339 29/01/ 2015). AI use statement Google Gemini was used exclusively to generate selected illustrative elements included in the Graphical Abstract. The generated images were reviewed and edited by the authors. No AI tools were used to generate, analyze, modify, or interpret the experimental data, scientific figures, or results presented in this manuscript. CRediT authorship contribution statement Meryem El Ouardi: Writing – original draft, Validation, Supervision, Methodology, Investigation. Isaac Vieco-Martí: Methodology, Investigation. Sofía Granados-Aparici: Methodology, Investigation. Inmaculada Noguera: Methodology, Investigation. Samuel Navarro: Writing – review & editing, Methodology, Investigation. Rosa Noguera: Writing – review & editing, Writing – original draft, Validation, Supervision, Investigation, Funding acquisition. Declaration of competing interest The authors declare that they have no competing financial or non-financial interests that could have influenced the work reported in this manuscript.