Patient-derived complex organoids are a critical tool to understand and explore pediatric malignant brain tumors

Abstract

Pediatric malignant brain tumors (PMBTs) remain among the most common and challenging cancers in children and adolescents, with current therapies often failing to deliver satisfactory outcomes. A major obstacle is their intrinsic and extrinsic resistance mechanisms, underscoring the urgent need for innovative therapeutic strategies and accurate preclinical modeling. Recent advances in organoid-based technologies offer promising tools to mimic PMBTs more faithfully in vitro. These three-dimensional (3D) models can replicate key features of the tumor and its brain-like microenvironment, providing valuable platforms for studying resistant cancer cells and testing novel treatment approaches. This review discusses the relevance of cultured 3D systems, organoids and tumoroids, in pediatric neuro-oncology, emphasizing their role in precision medicine. These models have become essential for dissecting the complex biology and dynamic biological processes of all PMBTs, while bridging clinical challenges with experimental discoveries, ultimately enabling more effective and personalized treatments for young patients.

Introduction

Pediatric central nervous system (CNS) tumors encompass an heterogenous group, spanning brain and spinal cord locations and multiple histotypes, as defined by the 5^th World Health Organization (WHO) classification of 2021 [1–3]. To decide appropriate therapies, based on the histomolecular refinements of this classification, pediatric malignant brain tumors (PMBT) are categorized into three main groups: embryonic tumors (such as medulloblastomas (MB) and atypical teratoid/rhabdoid tumors (ATRT)), gliomas (high- and low-grades, respectively HGG and LGG) and ependymomas (EPN) [1–6]. Each pediatric CNS histotype is also molecularly subdivided into several categories that reflect prognosis and metastatic risks. The identification of driving mutation or fusions in pediatric brain cancers played a pivotal role in their comprehension and became a powerful diagnostic tool now routinely used for those patients [1–6]. Nevertheless, to understand these drivers, their impact on tumoral cell ontogeny, and their role in intrinsic resistance or as potential therapeutic targets (Fig. 1) [7, 8], robust preclinical multi-omics approaches are required to accurately decipher the malignant specificities from RNA to protein modifications. Beyond epigenomic and genomic features used to classify PMBTs, the brain microenvironment provides critical insights into their pathogenesis, informed by developmental programs, tumoral imprinting of the normal brain cells and local immunity [9–14]. For example, posterior fossa tumors, which constitute over half of all PMBTs, are in correlation with patients’ young age and the putative origin of malignant cells [15]. These histological and biological disparities reflect inherent intra-tumoral cellular heterogeneity and microenvironmental interaction, underscoring PMBTs as fundamentally distinct entities. This complexity represents a major challenge for tackling treatment resistance, and improve clinical outcome and quality of life for those who survive the disease, but also emphasizes the urgent need for comprehensive models built with patient-derived cells and specific tissue bioengineering techniques (Fig. 1).

Fig. 1.

Applications of patient-derived 3D/organoid models. Patient-derived 3D/organoid models provide versatile platforms to investigate both fundamental (blue arrows) and translational (yellow arrows) aspects of tumor biology. Fundamental applications include exploring tumor origin (1), tumor-specific growth characteristics (2), and mechanisms of invasion and migration (3). Translational applications encompass drug screening (4), the study of treatment resistance (5), supporting preclinical evaluation of therapeutic responses and serving as a tool for precision medicine. Illustrations were created using BioRender.com

In the past, most modeling attempts have concentrated on constructing simple PBMT models deriving from primary cell lines and/or mouse models [16–21]. Recently, the field has seen major advances, thanks to the prominent rise of organoid culture in modeling both microenvironmental tissues and diseases [22], with the first cerebral organoids emerging a decade ago (Fig. 2A and B) [22, 23]. Although two-dimensional (2D) cell lines are easy to establish and rapidly available [16, 17, 20], three-dimensional (3D) modeling possibilities offer more faithful representations of PBMT’s by preserving native tumor heterogeneity, both at the molecular and cellular levels, and by including the brain microenvironmental interactions [24–27]. Simple structures like 3D-spheroids already provide more realistic representation of initial PMBTs [25, 26, 28–32]. Tumoroids, or tumoral organoids, represent more complex self-organized 3D multicellular structures and recapitulate the function, the tumoral 3D architecture, and the complexity of their native original tumor (Fig. 2A and B). Importantly, these models should represent the diverse cell types present in PBMT, notably including tumor cells, vascular endothelial cells, microglia, and immune populations, allowing their incorporation into organoid systems to more accurately capture tumor-microenvironment interactions. Recently, a nomenclature for those various modeling concepts has been proposed, aiming to enhance scientific communication and ensure more appropriate use of terms such as organoids, tumoroids, assembloids, and explants [33]. Importantly, these models enable the reduction, the refinement and the replacement of animal models (3R principle) [34, 35], while retaining a sophisticated model mimicking the diversity of PMBTs (Fig. 1). In line with the growing recognition of organoids as predictive, human-relevant models, the FDA Modernization Act 2.0, signed into law in December 2022, permits the use of alternative methods to animal testing, including organoids, for preclinical drug evaluation and validation. The European guidelines based on HYBRIDA project that are proposed to be integrated in ALLEA, which is designing the European Code of Integrity Conduct for Research, delineate the practical, ethical and intellectual challenges inherent to organoid research.

Fig. 2.

Overview of 3D patient-derived modeling technologies in pediatric brain tumors. A Schematic representation of various 3D modeling approaches and their relative complexity. Models range from simple spheroids to more complex organoids and tumor-on-chip systems. For each model type, the name, year of development, and a schematic depiction of the required components and culture conditions are provided. B Examples of multiple 3D- and organoid-based models derived from the same patient tumor sample (relapsed diffuse midline glioma, BT35). As illustrated by previously published results and preliminary findings generated by partners of the EN-HOPE SMART4CBT program, four distinct 3D models were established from identical patient-derived tumor cells: spheroids (Blandin et al., 2019), organoids, assembloids, and tumor-on-chip systems (Fuchs et al., 2021). Representative bright-field and fluorescent microscopy images depict differences in cellular organization and composition across models. Images for the engineered cerebroids and explant cultures fields are not included, as these specific models have not been established within our research center

EN-HOPE SMART4CBT center (East North-Hematology Oncology PEdiatric consortium offering research programs of Social sciences, Microenvironment & multi-omics Analyses in RadioTherapy resistance For Children Brain Tumors) recently labeled pediatric research center of excellence by the National Institute of Cancer in France grounded its biological program on the development of innovative organoid-based models to capture consistent and reproducible results to understand radioresistance in pediatric brain malignancies.

The hereby review, authored by the biological program task force of this research center, including both researchers and clinicians, highlights recent advances in non-animal 3D in vitro models across pediatric neuro-oncology. We focus on organoids, tumoroids, and related advanced models, emphasizing their production, expansion, preservation, and characterization, directly aligned with the landscape of pediatric brain tumors with the poorest outcomes. We discuss their unique features, current limitations, and balanced advantages and disadvantages, particularly in the context of studying all pediatric brain tumors and their related resistance to standard therapies. Our ultimate goal is to illustrate how these models are accelerating translational research and informing patient-centered therapeutic strategies.

PMBTs’ shared intrinsic and extrinsic specificities of both tumors and brain microenvironment

Understanding PMBT requires integrating their shared intrinsic molecular features and extrinsic microenvironmental influences, which together shape tumor heterogeneity, progression, and response to therapy. Subdividing histology-based entities into molecular subgroups has categorized patient populations into smaller groups, creating critical and challenging approaches when considering treatment by histological subgroup. Biological features that overlap phenotypes may unite tumors of disparate histologies and cells of origin and serve as therapeutic clues, thanks to the DNA, RNA and proteomics data available from translational research programs. Identifying similar omics features can then inform improved treatments across histologically distinct tumors. Recent publications [12, 13, 15, 36–38] indicate that three main PBMT histotypes (e.g., MBs, HGGs and EPN) share common signaling pathways and similar brain microenvironmental characteristics. A key shared hallmark is hypomethylation, detected by epigenetics, which reshapes chromatin organization and drives oncogenic transcription programs associated with more aggressive tumor cell phenotypes [36–38].

This epigenetic knowledge reveals a first level of intra-tumoral heterogeneity, arising from differential expression patterns superimposed on the various transcriptomic cell states identified in single-cell approaches. Indeed, all single-cell studies, from RNA sequencing to proteomics, including spatial analyses, reveal multiple cell populations within tumors, exhibiting both stem-like and proliferative capacities, along with a developmental hierarchy that connects the large undifferentiated cell population to the more aggressive histomolecular subgroups [12–15, 37, 38]. Furthermore, the microenvironmental cell type composition appears similar across most aggressive PMBTs, introducing confounding effects in molecular analyses. A key difference may lie in the relative cell type proportions (e.g., MB, EPN, glioma cells vs. non-tumoral cells) and how these compositions influence PMBT molecular profiles, cell dedifferentiation and tumor responses. It becomes then essential to investigate these factors experimentally, using innovative modeling that takes into consideration extrinsic characteristics, particularly the immune and neuronal landscapes. Notably, the prominent enrichment of myeloid cells across all these histotypes has been independently associated with PMBT prognosis and reflects the variability observed in cell-type proportions within tumor types [10–14, 39, 40].

In addition, recent publications from Monje’s team have demonstrated that local neuronal activity, shaped by the local brain development and the dedicated role of each brain region, contributes to tumor cell initiation and progression [41–43]. Furthermore, complex interactions between tumor and non-cancerous cells may be influenced by non-cellular characteristics, such as metabolite composition and oxygen availability, which are driven in part by vascular architecture and local brain matrix [17, 44, 45]. As 2D cultures recreating PMBT models cannot recapitulate these parameters, recent studies using 3D patient-derived in vitro models highlight the need to incorporate multiple levels of complexity (Fig. 1) [16, 17, 20, 29–32], consistently supporting the development of organoid-based systems [24, 26, 27, 44, 45].

The historical spheroids, the ancestor of the 3D/organoid-based conformational models and its evolution

Introduced as a first attempt to go beyond traditional patient-derived 2D cultures, spheroids represent the historical precursors of 3D conformational models, enabling the study of spatial heterogeneity and stem-like properties in pediatric brain tumor cells. Unlike neurospheres or tumor-stem cell culture systems composed of free-floating clusters of tumor precursors [17], spheroids (Fig. 2) involve self-assembly via cell–cell aggregation and adhesion, cultured in suspension [17, 30, 44, 46] or in a neutral matrices [29–32], preserving PBMT cell heterogeneity but lacking microenvironmental components. Patient-derived spheroid models (Fig. 1, 2A–B, Table 1) notably mimic hypoxic conditions in histone-mutant pediatric gliomas and replicate key characteristics of parental tumors [17]. These spheroids feature the same proliferative and renewal capacities as the original tumor samples, while preserving stem-like characteristics. A similar approach has been developed for Sonic Hedgehog (SHh) pediatric MBs, where researchers employ genetically engineered mice to initiate MB tumors [24, 29, 47, 48]. After isolating mouse MB cells and dissociating them into single-cell suspensions, cells were cultured either in the traditional 2D manner or in a matrix-free suspension, forming spheroids. These 3D structures reach an average size of approximately 100 µm within 2 to 4 days and recapitulate the multiple cell passages observed in tumors (Table 1). They can also be cryopreserved to ensure experimental consistency and accelerate in vitro research [24, 25, 29, 31, 32]. Several studies report spheroid growth in serum-free media, preserving spheroid tumorigenicity, as evidenced by successful xenograft experiments, unlike cells from 2D cultures [24], or by initiating subsequent both 2D cultures and neurospheres [17, 24, 25]. This type of tumor miniatures are widely applicable in various translational and biomedical fields, such as studying tumor initiation and invasiveness and drug screening (Fig. 1).

Table 1.

Comparisons of the different 3D modelling techniques

	Spheroids	Cerebroids	Explant cultures	Tumoroids	Assembloids	Tumor-on-chip
Advantages	Limited cost Easy to use/produce Reproducibility Limited experiments	Cell differentiation Engineering of genetic modifications Enables complex experiments	Easy to initiate Representative of original sample Limited experiments	Heterogeneity preservation Representative of original sample Enables complex experiments	Tumoral cell culture Integrate microenvironment Enables complex experiments	Integration into microfluidic flow Allows entry/exit of cells of interest Facilitates refreshing of culture medium
Disadvantages	No cellular heterogeneity Not matching original tissue	High cost Time-consuming Complex medium culture	Rare starting material No long-term cell culture (only 1 or 2 passages) Complex medium culture	Rare starting material High costs Time consuming Complex medium culture	2D or 3D tumor cell culture High cost Time-consuming Complex culture	Rare starting material Requires specific microfluidic expertise High costs Time consuming Complex culture media
Time of development	2–4 days	1 to 2 months	Few weeks	6 weeks to 2 months	1 to 3 months	6 weeks to 2 months
Size	300 to 1000 µm	1000 to 4000 µm	1000 to 4000 µm	500 to 2000 µm	300 to 4000 µm	500 µm to 2 mm

The EN-HOPE SMART4CBT center has developped this type of model, delineating spheroids to explore environmental invasiveness and evaluate hypoxic impact on tumor cell–cell interactions (Fig. 2B) [17].

More sophisticated spheroid-like models (Fig. 2A and B) have been generated directly from fresh tissue samples or patient-derived xenografts (PDX) [49]. After mincing specimens into small pieces, tumoral tissues were seeded without single-cell dissociation into agar-coated flasks or low-adhesion plates, together with organoid medium [49–53]. Small spheroid-like structures form within two weeks, with a size ranging from 300 µm to 1000 µm, more closely resembling explants (Table 1). These 3D cultures may be transfected or retransplanted into animal brains to generate organoid-derived xenografts. EPNs have also benefited from this approach to monitor tumor cell interplays within scaffolds and examine microenvironmental regulators during tumor growth [54, 55]. Scaffolds or extracellular matrix gels (ECMG) help to maintain relevant 3D conformational growth, with histotype-specific preferences for materials such as Matrigel or collagen [25, 54, 55]. Complexified spheroid models pave the way toward investigations on tumor cell–cell aggregation and interactions in growth factor-free culture conditions and can even be cultured without any matrix [56]. Nevertheless, specific scaffolds or matrices remain useful to recreate the brain cell microenvironment, bringing these models closer to cerebroid systems.

Cerebral organoids or cerebroids

Beyond spheroid-based models, the initial development of brain organoids focused on non-tumoral cerebral organoids (cerebroids), which were established to recapitulate normal human brain development and tissue organization before being adapted to model brain tumors. Bona fide organoids (Fig. 2A and B) are defined as self-organized 3D structures derived from stem cells or progenitors (pluripotent, fetal, or adult), characterized by self-renewal and differentiation properties. Organoids phenocopy, at least in part, the functional, structural, and biological complexity of the organs or tissues [22, 23, 27, 56, 57] (Fig. 1). Stem cell self-renewal ensures viability, reproducibility and long-term maintenance.

The production of brain organoids or so-called cerebroids necessitates four fundamental components, namely stem cells, matrix, soluble factors, and physical cues. Cerebroids can originate from induced pluripotent stem cells (iPSCs), embryonic stem cells (ESCs), or adult stem cells [22, 23, 58–64]. Two major methodological strategies can be commonly used for brain organoid formation: non-directed and directed differentiation.

In non-directed approaches, as initially described in the standard method developed by Lancaster et al. [22, 23], iPSCs or ESCs are cultured in suspension to generate stem cell clusters named embryoid bodies (EBs) [59, 60] through mechanical or enzymatic dissociation. These EBs recapitulate early stages of embryogenesis and can differentiate into the three germ layers (endoderm, mesoderm, and ectoderm) [59, 60, 65, 66]. Upon embedding in an extracellular matrix mimicking the brain microenvironment, neural induction and regional specification emerge through endogenous signaling gradients. This strategy allows the generation of heterogeneous and spatially complex brain-like tissues, capturing aspects of early human neurodevelopment and cellular diversity, but often at the cost of increased variability between organoids and limited control over regional identity.

In contrast, directed differentiation approaches rely on the stepwise application of defined soluble factors and small molecules to bias lineage commitment and regional specification toward specific brain identities (e.g., forebrain, midbrain, cerebellum) [59, 60]. By modulating key developmental signaling pathways, these methods improve reproducibility and enable more consistent generation of region-specific organoids. However, directed approaches may restrict self-organizing capacities and reduce the emergence of complex cytoarchitectures, potentially limiting their ability to model global brain development and long-range tissue interactions [59, 60]. The choice between directed and non-directed strategies therefore depends on the biological question addressed, balancing experimental control against developmental complexity.

Following EB formation, organoids are embedded in a matrix recapitulating the native brain environment to support further neural differentiation and cerebroid maturation. Several cell-free extracellular matrices can be used, including basement membrane extracts [67], Matrigel [68], synthetic hydrogels [68], or collagen [69]. Earlier Matrigel formulations exhibited batch-to-batch variability, but contemporary preparations are better defined, improving reproducibility. Alternative matrices may be fully synthetic and/or custom-made, allowing adaptation for each specific experimental model [68, 70]. Importantly, extracellular matrix (ECM) composition might vary with aging, tumoral process and stem cell culture [65, 71].

Metalloproteases remodel the ECM by degrading and reorganizing its components, which can indirectly contribute to matrix stiffening in some tumor modeling, in conjunction with increased ECM synthesis. PBMT’s ECM is usually more enriched in proteoglycans and collagen than its adult counterpart [65, 70, 71].

Cerebroids within matrixial environment can reach diameters of 1000 to 4000 µm and need 1 to 2 months to mature (Table 1). Cerebroid cells, recently developed in our research centers for drug testing (examples shown in Fig. 2B), can be cryopreserved, and their structure analyzed using conventional and immunofluorescence microscopy. To facilitate and support the development of these large and highly specialized organoids, optimized culture media supplemented with appropriate growth factors are needed to induce proper differentiation, activity, and cellular communication, and must be adjusted throughout the culture steps [22, 23, 68–73]. Soluble factors and small molecules are delivered in a controlled manner to influence cell–cell interactions and modulate signaling pathways, ensuring reproducible developmental trajectories. Beyond molecular cues, recent insights on incorporating microglial cells or specific electrical neuronal activity aim at addressing their role in cerebroid development and maturation (Fig. 1) [15, 26, 43, 44, 46, 64].

Engineered brain tumor cerebroids

Derived from the standard cerebroid method, genetically engineered organoids have emerged as powerful tools to investigate tumor initiation and its early oncogenic events. Genome-editing approaches (Fig. 2A) allow to test at any step of PMBTs’ development the tumorigenic capability of gain- and loss-of-function mutations or to understand the impact of copy number variations in several tumor co-drivers. These strategies are compatible with both directed and non-directed organoid formation protocols, although directed approaches may offer improved regional specificity when modeling tumors arising from defined neurodevelopmental lineages. Bian et al. have generated neoplastic cerebral organoids from classical cerebroids [22, 23, 61, 65], by introducing defined oncogenic mutations into developing brain cells [74]. Employing molecular tools such as CRISPR-Cas9 and Sleeping-Beauty transposon-mediated gene insertion, they recreated multiple isolated mutations or amplifications, as well as combined oncogenic programs seen in both adult and pediatric brain tumors [74]. Pediatric tumor–relevant alterations included H3.3K27M or SMARB1 mutations, as well as multiple passenger alterations in p53, CDK4, CDKN2A, MDM2, or PDGFRA. In the same way, other publications have pushed forward those strategies in multiple histotypes of PMBTs [25, 74–81]. In most cases, genetic alterations were introduced before matrix embedding, allowing tumorigenic clones to expand during 3D organoid growth. Automated 3D proliferation monitoring enabled quantitative assessment of tumor expansion using specific protein markers [25, 74–77, 79].

In pediatric high-grade glioma (pHGG) or in MBs [22, 25, 74–78], engineered organoids successfully captured molecular heterogeneous tumor subgroups and translate accurately genome-wide screening signatures into tumor generation processes. For example [25, 76, 77], previously key genetic drivers of MB subsets were genomically transferred into cerebellar organoid cells based on gene-modulation combinations of c-MYC and Otx2, or c-MYC and Gfi1 overexpression, as well as PTCH1 mutation [25, 76, 77]. Utilizing CRISPR-Cas9 or transposon system, the research teams have successfully replicated the specific subsets of MBs, demonstrating the importance of neurodevelopmental and epigenetic molecular deconvolution to generate a given MB subgroup. Similarly, H3.3K27M mutations were incorporated into iPSCs before or during EB formation, aiming to create diffuse intrinsic pontine glioma (DIPG) organoids, and thereby deciphering the role of initiating oncogenic events [74, 75]. Those studies also proposed to further characterize the role of gene-edited organoids by orthotopic injection into mice.

Based on the same methodology, Ogawa et al. [82] modeled adult glioblastomas (GBM) by introducing concomitantly TP53 locus modulation and HRasG12V expression into 4-month-old cerebroids via electroporation and generating GBM-like organoids within 8–16 weeks. As for organoid-PDX, GBM-cerebroids were dissociated and cultured as neurospheres or co-cultured with non-tumoral cerebroids to explore the invasion capacities of GBM cells [74, 78, 79, 81, 82].

More recent insights take advantage of 3D bioprinting technologies as a powerful tool to generate engineered PMBT-cerebroids [82] with improved spatial control and reproducibility. Collectively, these studies highlight engineered organoids as essential tools for deciphering major molecular and epigenetic signatures defining each PBMT histotypes and subgroups (Fig. 1). PMBT model engineering remains unsuccessful in the EPNs, where molecular classification is not based on driver mutations or gene deregulation. Unlike strict tumoroids or patient- derived tumor organoids, engineered cerebroids are not fully recapitulate intra-tumoral heterogeneity. These shortcomings call for more complex models called assembloids, which mix cerebroids with cancer cells or tumoroids (patient-tumor-derived organoids) [33].

Cerebroid-tumor cell coculture (assembloids) as an invasion model

To go further and incorporate the brain microenvironment into PMBT cell cultures, recent studies have cocultured cerebroids with tumor cells, aiming to study tumor complexity, modifications based on microenvironment and tumor cell invasiveness (Fig. 2) [33, 44]. A recent publication on the generation of microglia-containing brain organoids explored the morphological and migratory characteristics of microglia in the PMBT context, using multiple neuronal subtypes to mimic the brain tumor microenvironment [44]. These microglia-containing organoids were fused with spheroids derived from H3K27M-mutant pHGGs bearing notably TP53 mutations to recreate diffuse midline-like locations. This model permitted studies on tumor cell-microglia interactions and invasion within a few weeks, promoting the development of such composite systems.

In parallel, a larger body of literature has leveraged complex assembloids in adult brain tumor setting [78–82]. To mimic GBM cell invasion in adult brain tumors, Linkous and colleagues cultured GFP-labelled GBM stem cells (GSCs) with cerebroids to generate brain organoid gliomas [78]. Remarkably, after only one week of co-culture, they were able to closely follow growth and invasion of tumor cells into brain organoids. The tumor cells that invaded non-tumoral cerebroids exhibited sustained proliferative capacity. To confirm whether this resulted from the interaction with the brain microenvironment rather than with the culture media, the authors compared cerebroid differentiation medium with GSC medium. GSCs exhibited a 70% reduction in proliferation when cultivated in the cerebroid differentiation medium, suggesting that the enhanced proliferation observed in this model resulted from direct interactions between GSCs and the brain microenvironment. The study further identified gap junction-mediated interconnecting microtubes and its complex network paving physical routes for GBM cell invasiveness into cerebroids. Similarly, Krieger et al. [79], as well as Da Silva et al. [74, 83], generated cerebroids from iPSCs and cultured them with GBM cells. They were able to use this system for brain invasion studies, confirming the highly invasive properties of spheroids derived from human GBM cells. These hybrid organoids usually formed within a few days and had relatively large sizes (300–4000 µm) [83, 84]. Importantly, adult-focused studies [85, 86] demonstrated that the tumor microenvironment is critical for maintaining the cellular states observed in primary glioblastomas. While highly informative, such adult GBM assembloid models operate in a biological context that lacks ongoing neural development, underscoring fundamental differences from pediatric brain tumor systems. In fact, pediatric brain tumors arise within a developmental and glial landscape, where tumor-microenvironment interactions are shaped by active neurodevelopmental programs [84]. In a recent publication, proteomics helped revealing microenvironment-induced reprogramming in DIPG assembloids, highlighting the specificity of pediatric disease that cannot be fully extrapolated from adult models. Overall, assembloid models demonstrate that brain organoids are relevant platforms to study tumor invasion in both adult and pediatric brain tumors, using three main approaches: a normal supporting tissue combined with established tumor cells (e.g., PMBT spheroids or explants) or incorporating PMBT cells during the process of cerebroids differentiation from iPSC. We illustrate in Fig. 2B one example of our recent iPSC derived cerebroids compatible to co-culture with 3D patient-derived cell lines. These models are embedded in Matrigel and cultivated using conventional brain organoids, often termed “hybrid organoids” [80], and provide a relevant support to explain closely PMBT invasion and their microenvironmental interactions (Figs. 1 and 2B) [81].

A variety of PMBT tumoroids

Tumoroids are defined as organoids derived from patient tumors [33]. They are generated by embedding either dissociated cells (obtained through mincing and/or enzymatic digestion of tumor tissue) or small intact tissue fragments (referred to as explants) [87] from fresh tumor samples into a matrix-based dome. This process supports the formation of three-dimensional, self-organizing structures that faithfully recapitulate the tumor cellular and spatial heterogeneity of the original patient sample (Table 1 and Fig. 2). As such, tumoroids provide biologically relevant models that bridge the gap between 2D cultures and in vivo systems.

For in-depth studies of tumoral processes, PMBT tumoroids (tumor cell-organoids) have been developed by adapting protocols initially established for adult brain tumors [87–89]. Paassen et al. generated tumoroids using both patient-derived orthotopic xenografts and fresh biopsies of ATRTs [90]. Tumor samples were minced and cultured either in suspension or embedded in a matrix, favoring cell survival, proliferation and differentiation. The resulting tumoroids, obtained within 1–2 weeks, retained ATRT subgroup-specific DNA methylation landscapes and transcriptional programs. The derivation efficiency was subgroup-dependent, with a 60% rate of success for ATRT-SHH (ATRT-sonic hedgehog) and 50% for ATRT-MYC [90]. Multiple studies have reported the successful establishment of tumoroids from MBs [24, 25, 76], EPNs [24, 25], pHGGs and LGGs [25–27], generated at diagnosis or relapse. Two main derivation methods have been systematically compared to determine optimal conditions. One approach involved total cell dissociation followed by spheroid aggregation, while the other, more similar to explant cultures, used small tumor fragments (0.5 to 2 mm) seeded directly into a matrix. Those so-called patient-derived organoids constitute robust and reproducible model that can be generated within 6 weeks to 2 months. They accurately preserve key features of the parental tumors, including mutational profiles and DNA methylation patterns, but also capable to reinitiate tumor growth after injection into mice (Figs. 1 and 2, Table 1) [25]. Models of DIPG and diffuse midline gliomas (DMGs), often established from tumor-derived glioma stem cell populations, have represented a major breakthrough for investigating invasion processes, cellular plasticity, and therapeutic vulnerabilities [25–27, 75, 77, 83, 84]. In our own research center, those innovative 3D tumoroid models of pHGG, MBs and EPNs are now used to investigate the microenvironmental influences and how the cells-to-cells interaction, hypoxia, and the dynamic processes by which tumors adapt and become resistant to irradiation (Fig. 2B).

In previous publications on adult brain cancers [86–89], procedures relying on fresh tumor samples are highly successful at generating tumoroids that reproduce the cellular and molecular landscape of adult glioblastomas. In pediatric setting, the “explant-like” process was the easiest and fastest way to obtain matrixial organoids. Both models as in pediatric organoids were able to explore microenvironmental influences. Complementary to these works, Jacob et al. [88, 89] generated tumoral organoids from fresh glioblastoma samples cut into 1-mm pieces within 2 weeks, thanks to permanent agitation in ultra-low attachment plates. With the exception of ATRT, organoid generation from both adult and pediatric brain tumors consistently achieved success rates that exceeded 85%, enabling the establishment of large and representative collections encompassing the majority of clinical and molecular tumor subtypes [25–27, 75, 77, 85–90]. These small “tumors-in-a-dish” can reach up to 2–4 mm and may recapitulate the tumor cell microenvironment, including partial tumor vasculature and/or immune cells. Nevertheless, this model still lacks most of the microphysiological features that constitute the brain microenvironment.

Tumor-on-chip: complexification through microfluidics

So, the so-called on-chip technologies have emerged as an alternative tool capable to recapitulate a microphysiological environment with 3D structures resembling in vivo organs. Such bioengineering, involving microfluidic channels and 3D cell cultures, facilitate the integration and compartmentalization of endothelial, neuronal, microglial, or immune cells under physiological or pathological conditions (e.g., hypoxia, metabolic rewiring or cytokine/chemokine trafficking) (Fig. 2A and B) [91–95]. These devices help to unravel different aspects inherent to tumor microenvironment. Indeed, vessel functionality and blood flow viscosity influence cell mobility into or out of the tumor vasculature, but also in neoangiogenic sprouting, which results in metabolic and oxygen tumoral gradients within the tumor.

Tumors-on-chip also offer to trace the production and effects of pro-angiogenic factors and can mimic both normal and disrupted blood–brain barrier [93, 95]. The recreation of a physiological tumor vasculature is instrumental for studying immune cell infiltration. This invasive process needs several steps, including haptotaxis via gradients of adhesive cues, adhesion driven by pro-inflammatory signals, and transmigration across the endothelial barrier to reach the tumor core. Beyond the vasculature and based on the recent insights into the relationship between pHGG cells and neuronal activity [11, 13, 41–43], neurons and their circuitry can be also modeled by microfluidic chips. Of particular interest, these bioengineered systems may enable electrophysiological recordings and the tracking of neuro-glial interactions. This is exemplified by our own work shown in Fig. 2B, where we demonstrate our ability to culture in such “on-chip” devices patient-derived 3D cell models (e.g., spheroids and neurospheres) [95]. By integrating microelectrode arrays into the compartmentalized microfluidic device, the electrophysiological activity of glutamatergic neurons could be easily assessed and was significantly increased in presence of malignant DIPG cells, underlining the neuron-tumor cell interactions and their crosstalk. Multiple devices have been recently developed to envision and quantify brain-malignant cell networks, while also enabling the concomitant application of therapeutic strategies such as irradiation or targeted therapies. Altogether, tumor-on-chip technologies represent a major step forward in the modeling of pediatric brain tumors by their capacity to integrate vascular, immune, and neuronal dimensions within a single, controllable platform. These systems provide a powerful framework to advance our understanding of cellular networks, trafficking, and adaptive responses within the tumor microenvironment, while also offering promising perspectives for the development of personalized therapeutic strategies (Figs. 1 and 2).

The role of growth factors in organoids and assembloids

The in vitro culture medium composition and supplemental growth factors play a crucial role in the development, growth, and long-term maintenance of stem cell and 3D cultures [96], leading to the recent advances in those organoid-based modeling. Across all publications there are no specific variations observed in the culture medium composition used for adult vs. pediatric models. In both contexts, tumor cells are most commonly derived in neurobasal medium or in combination with DMEM, enriched with components essential for cellular metabolism and homeostasis, including cobalamin (vitamin B12), alanine, zinc sulfate, HEPES buffer, and sodium pyruvate [17, 22, 23, 27, 96–100].

In most studies, if not all, cultures are further supplemented with growth factors, such as epidermal growth factor and basic fibroblast growth factor, which promote the survival and expansion of neural stem-like and progenitor cell populations. B27 and N2 supplements are almost universally incorporated, as they provide antioxidants, lipids, hormones, and micronutrients essential for the survival of primary brain cells in long-term culture [22, 23, 27, 101, 102]. Insulin is also systematically added to media, reflecting its central role in metabolic regulation and growth signaling. While these supplements are indispensable for sustaining in vitro cultures, they also exert strong selective pressures that may favor specific cellular states over others. Indeed, growth factor–enriched conditions are not biologically neutral and can introduce significant biases by promoting stem-like, proliferative, or undifferentiated phenotypes at the expense of more differentiated or quiescent tumor cell populations. Such biases may alter the relative abundance of cellular subclones, reshape lineage hierarchies, and influence transcriptional and epigenetic states within organoid and tumoroid cultures. As a consequence, growth factor composition can directly impact cellular plasticity, enabling tumor cells to dynamically switch between proliferative, stem-like, and differentiated states, These culture conditions not only shape baseline tumor biology but may also precondition tumoroids toward adaptive responses that influence radiation sensitivity and resistance mechanisms observed in vitro. Additional supplements can further modulate these effects. Agents that stabilize the culture environment and limit oxidative stress, such as β-mercaptoethanol, as well as lineage-specific growth factors that promote neuronal or glial differentiation within organoids and tumoroids [22, 23, 27, 72, 90]. However, such refinements also increase protocol variability and complicate cross-study comparisons, particularly when assessing treatment responses.

To date, no consensual guidelines exist regarding the optimal combination or concentration of growth factors for brain tumor organoid or tumoroid cultures. No formulation has yet reached a sufficient level of confidence to be adopted systematically across laboratories worldwide. In our own experience, robust evidence demonstrating that specific growth factor cocktails consistently improve tumoroid derivation efficiency or faithfully preserve treatment-relevant phenotypes, particularly with respect to radioresistance, remains limited. This highlights the need for standardized benchmarking studies explicitly evaluating how culture conditions influence plasticity and therapy response.

Another major challenge will be linked to the assembloid systems, which will require carefully balanced growth factor cocktails capable of supporting multiple cellular compartments without introducing dominant biases that would modify intrinsic tumor behavior, particularly in the study of adaptive resistance to irradiation.

Models’ validation strategies (Figs. 2 and 3)

Fig. 3.

Validation workflow for patient-derived tumor models. Validation of tumoroids requires comparison with the corresponding patient tumor, ensuring preservation of 3D structure, cellular heterogeneity, and microenvironmental features. Morphological assessments (hematoxylin/eosin colorations, immunohistochemistry) and high-resolution 3D imaging (immunofluorescence with clearing) evaluate structural and phenotypic fidelity. Multi-omics approaches, RNA/exome sequencing, DNA methylation profiling, and spatial/single-cell methods, assess molecular fidelity and heterogeneity. Integrating these complementary methods with clinical data provides a robust strategy to validate tumoroid accuracy. Illustrations were created using BioRender.com

When developing tumor models, the essential criteria at medical and preclinical point of view is to compare in vitro models with the patient-derived cells and to ensure a relevant 3D structure of the pediatric brain tumors, as well as the preservation of their cellular and microenvironmental heterogeneities. An ideal model should retain brain regional features, such as microglia, the blood brain barrier, and vascular structures, along with precise characterization of both physiological and tumor cell populations, usually based on spatial and multi-omic profiling. No consensus guidelines have been established to guide model development, but numerous methods exist for their validation. Each method may be complementary and can be directly linked to paired clinical data, including imaging. The first step in organoid/tumoroid validation should involve morphological assessments.

To describe the cellular composition of a tumoroid, structural information obtained from hematoxylin/eosin colorations (H&E) is complemented by immunohistochemistry (IHC), which targets several tumoral biomarkers. These include diagnostic markers (e.g., driver mutation staining and DNA methylation-related markers) [5, 17, 21, 25–27, 76–78] and cell identifiers (e.g., Nestin or CD98 for neural progenitors, GFAP (Glial Fibrillary Acidic Protein) and β3-tubulin for assessing glial and neuronal differentiation (examples shown in Fig. 2B) [27, 103, 104]. IHC on 3D models involves paraffin embedding, fixation, dehydration, slicing, and deparaffinization steps prior to antibody staining. While H&E and IHC are fast and easily set up techniques, staining sections of a tumoroid does not permit precise visualization of cell-to-cell structures and multiple concomitant staining. Consequently, immunofluorescence is extensively used, which notably allows high-resolution 3D imaging by light sheet microscopy associated with tissue-clearing methods [25, 105]. This phenotyping technology of the entire organoid/tumoroid preserves the 3D architecture, while allowing the detection of multiple molecular targets. However, the clearing step remains time-consuming and may lead to material loss. Complementary to morphology-based assessments, when considering cerebroids and/or assembloids, study of electrical activity, as illustrated in Fig. 2B, together with analyses of neurotransmission trafficking, helps understanding the role of neuronal modularity and its projections to brain structures. Adding to the morphological evaluation, sequencing approaches of the whole tumoroid (e.g., RNA or exome sequencing, DNA methylation profiling, microsatellite and copy number variations assessments), either alone or combined with spatial methodologies or single-cell isolation, are crucial for in-depth analyses [25–27, 44, 45, 77, 79, 105–111]. The integration of innovative tumor-derived multi-omics into organoids, especially spatial omics mapping [11, 12, 14, 24, 25, 27, 38, 108–110, 112], enables the sequencing of tissue sections and dynamic spatial mapping to address fundamental questions related to tumoroid heterogeneity and its correspondence to the related paired tumor. No real consensus is available on the optimal combination of omics approaches to validate tumoroids, but this implies at least to define the main mutational events or genomic rearrangements present in the paired pediatric tumors. National and international initiatives centered on tumor diagnoses [1–4] open the path to explore genetic changes and epigenetic alterations and investigate the preservation of DNA methylation. In 2020, the Human Cancer Models Initiative proposed to compare side-by-side the methylation landscape of organoids and 2D-adherent models from various tissues (colon, pancreas, stomach, lung, and esophageal tumoral specimens) [107], leading to an increasing number of studies adopting this methylation-based validation method. Overall, an integrative approach that combines multiple complementary methodologies represents the most robust strategy to validate the accuracy and fidelity of individual tumoroids.

Real-world and routine applicability of tumoroid-based models in precision medicine

Beyond their value as experimental systems, tumoroid and related 3D models increasingly demonstrate their relevance in real-world and routine clinical–research settings, particularly in the context of precision medicine [25, 27]. Unlike genetically engineered models or long-term cell lines, patient-derived tumoroids can be established directly from diagnostic or relapsed tumor material within clinically meaningful timeframes that may be compatible with therapeutic decision windows [25, 49, 75, 87]. This accessibility positions tumoroids as translational tools capable of bridging molecular diagnostics and functional precision oncology (FPO), especially in rare and heterogeneous pediatric brain tumors for which evidence-based treatment options remain limited.

In routine settings, tumoroids will offer a scalable and adaptable platform that may be integrated into existing diagnostic workflows alongside histopathology, imaging, and molecular profiling. Indeed, their capacity to preserve patient-specific genetic, epigenetic, and phenotypic features will enable functional testing of standard-of-care treatments, including radiotherapy and chemotherapy, as well as targeted agents selected on the basis of molecular alterations [24–27, 77, 84, 90]. Importantly, several studies have shown that drug responses observed in tumoroids correlate with clinical outcomes, supporting their potential use as predictive tools rather than purely exploratory models [25, 75, 79, 110]. It will also serve as a representative model to better understand relapse mechanisms and to propose new therapeutic strategies at this critical stage.

From a real-world perspective, the flexibility of tumoroid platforms will be a major advantage. Tumoroids can be expanded into living biobanks representing diverse molecular subgroups, enabling retrospective analyses, cross-patient comparisons, and continuous refinement of predictive assays as new therapeutic options emerge [25–27, 75, 77]. Such biobanks are increasingly aligned with national and international initiatives aiming to harmonize molecular diagnostics and FPO across centers [1–4, 25, 27].

Crucially, tumoroid-based approaches complement, rather than replace, genomic-driven precision medicine. While molecular profiling identifies actionable alterations, tumoroids will potentially provide a functional layer that captures tumor cell plasticity, microenvironmental influences, and adaptive resistance mechanisms that are not fully predictable from genomics alone. This will be particularly relevant for radiotherapy, where intrinsic and acquired radioresistance are shaped by cell-state dynamics, hypoxia, and DNA damage response pathways [17, 25, 28, 41]. In this context, tumoroids enable iterative testing of treatment combinations and dosing strategies, offering a pragmatic framework for FPO.

Altogether, the integration of tumoroid models into the clinical context represents a realistic step toward individualized treatment strategies in pediatric brain tumors. By combining feasibility, biological fidelity, and adaptability, these models will support a system in which patient-derived data continuously inform experimental modeling and, conversely, functional assays refine therapeutic choices, fully aligned with the principles of precision medicine.

Ethics consideration and biobanking

As with other experimental tumor models, the development and implementation of organoid-, tumoroid-, and assembloid-based platforms require robust ethical oversight and well-structured biobanking infrastructures. Researchers and clinical institutions, including our own research center, have to ensure storage methods adapted to each type of biological samples and support the development of biobanking embedded within national or international initiatives, such as the Groupement De Recherche sur les Organoïdes, a network that unites more than 300 research teams and 14 production platforms in France, or comparable consortia worldwide. The quality control policy, sample traceability, and secure linkage to clinical and biological associated data are central to underpin organoid reproducibility, data integrity and long-term translational value. In addition, biomedical research involving patients’ samples must comply with ethical principles, regulatory requirements and informed consent procedures. At the European level, the HYBRIDA consortium worked specifically on organoid research, providing useful recommendations to improve ethical transparency toward patients, peers, evaluators and the general public. This is particularly critical in pediatric oncology, where ethical sensitivity is heightened by patient vulnerability, limited tissue availability, and long-term data governance considerations. These HYBRIDA initiatives are grounded in the ALLEA Code of Conduct for research integrity, emphasizing reliability, honesty, respect, and accountability, as well as in the 2021 WHO expert group recommendations on governance of human genome editing. Within this framework, three essential tools have been developed: MIAOU (Minimal Information About an Organoid and its Use for researchers), which promotes standardized reporting; ECHOES (Evaluator CHecklist for Organoid Ethical Studies), which supports ethical assessment; and RICOCHECK (Research Integrity Committee Organoid CHECKlist), which assists institutional oversight. Together, they provide all multifaceted features of organoid evolving research and represent an essential framework to guide and implement a responsible and reproducible research on organoids. They encompass and detail the ethical guidelines to obtain patient-derived tissues and their subsequent generated data, encouraging standardization of methods and the purposes of the organoid research that will be carried out.

In parallel, regulatory landscapes are evolving to explicitly recognize the translational relevance of human-relevant, non-animal models. Notably, the FDA Modernization Act 2.0 (2022) represents a landmark shift by permitting the use of alternative preclinical models, including organoids, assembloids and other microphysiological systems, with the goal of reducing animal testing for drug development and safety evaluation. This regulatory evolution directly supports the integration of patient-derived models into translational pipelines and aligns with global efforts to improve the clinical predictivity of preclinical research while reducing reliance on animal models.

Collectively, these ethical and regulatory frameworks strongly support the implementation of organoid-based FPO, particularly in pediatric brain tumor research to ultimately improved children’s outcomes.

Conclusions

The progress made in cell culture methods over the past two decades, combined with emerging technologies, has led to the development of innovative 3D models (Table 1) that more accurately mimic the complexity of pediatric brain tumors. Many of the earliest organoid-based models focused on pHGGs, particularly entities with extremely poor prognosis, reflecting an urgent unmet clinical need. These models have since become essential tools for investigating biological processes, replicating both normal and pathological in vivo scenarios, and unraveling the complex mechanisms and specific microenvironments of all aggressive pediatric brain cancers, not only pHGGs. This organoid technology has the potential to significantly reduce reliance on animal models (e.g., mice, rats, rabbits, and zebrafish), especially in the context of growing public concern regarding animal welfare. Moreover, organoids offer now new opportunities to investigate 3D structural organization more rapidly, while integrating pediatric tumor complexity. The expansion of organoid technology also brings new challenges, including high costs, longer processing times compared to conventional 2D cultures, and the current lack of consensus regarding validation methods.

Numerous protocols are currently being developed, ranging from engineered organoids and tumoroids to co-culture systems incorporating tumor cells and microphysiological platforms. As we face increasingly complex biological and preclinical questions in pediatric brain cancer research, organoid-based technologies have become indispensable in all histological entities. The development and use of such 3D tumor avatars are now essential and unavoidable.

Author contributions

M.D. and N.E.W. conceptualized the review and N.E.W. provided supervision. M.D., S.M. and N.E.W. wrote initial draft and all other authors added complementary sections and modifications. M.D., N.E.W., C.S., H.C. and E.C. provided preliminary data and pictures from the EN-HOPE SMART4CBT projects. NE.W., A.C., A.J., M.A.K., P.C., S.R., G.N. and H.S.B. are providing the patient data and samples. M.D., A.V., S.M., F.A., H.B., M.S., Ma.Du., A.R.R., C.B., S.P., and N.E.W. are the task force for organoid/tumoroid generation. M.D. and C.S. made the figures and tables. All authors have read and agreed to the last version of the manuscript.

Funding

The organoid development initiative and the conduct of this work were supported by the French National Cancer Institute (INCa) through the Pediacriex labeling of the EN HOPE SMART4CBT center. Previous work on microfluidic modeling was supported by a grant from the Ligue contre le Cancer.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.