Stem cell-derived neural organoids as platforms to investigate glioblastoma invasion and migration: A systematic review

Abstract

BACKGROUND

Glioblastoma multiforme (GBM) is the most aggressive and prevalent primary malignant brain tumor in adults, marked by poor prognosis and high invasiveness. Traditional GBM invasion assays, such as those involving mouse brain xenografts, are often time-consuming and limited in efficiency. In this context, stem cell-derived neural organoids (NOs) have emerged as advanced, three-dimensional, human-relevant platforms that mimic the cellular architecture and microenvironment of the human brain. These models provide novel opportunities to investigate glioblastoma stem cell invasion, a critical driver of tumor progression and therapeutic resistance.

AIM

To evaluate studies using stem cell-derived NOs to model glioblastoma migration/invasion, focusing on methodologies, applications and therapeutic implications.

METHODS

We conducted a systematic review following PRISMA guidelines, searching PubMed and Scopus for studies published between March 2019 and March 2025 that investigated NOs in the context of glioblastoma invasion/migration. After screening 377 articles based on predefined inclusion and exclusion criteria, 10 original research articles were selected for analysis. Extracted data were categorized into four analytical domains: (1) Tumor model formation; (2) NO characteristics; (3) NO differentiation protocols; and (4) Invasion/migration assessment methodologies.

RESULTS

The included studies exhibit significant methodological heterogeneity GBM model development, particularly regarding model type, cell source and culture conditions. Most studies (70%) used suspension cell models, while 30% employed spheroids, with most research focusing on patient-derived glioblastoma stem cells. NOs were predominantly generated from human induced pluripotent stem cells using both guided and unguided differentiation protocols. Confocal fluorescence microscopy was the primary method used for assessing invasion, revealing invasion depths of up to 300 μm. Organoid maturity and co-culture duration influenced results, while key factors for model optimization included tumor cell density, organoid age and extracellular matrix composition. Some studies also tested therapeutic strategies such as Zika virus and microRNA modulation. Collectively, findings support the utility of NOs as effective tools for studying GBM behavior and therapeutic responses in a humanized three-dimensional context.

CONCLUSION

Human NOs represent promising platforms for modeling glioblastoma invasion in a humanized three-dimensional environment. However, a limited number of studies and methodological heterogeneity hinder reproducibility. Protocol standardization is essential to enhance the translational application of these models.

Key Words: Glioblastoma; Stem cell; Organoid; Spheroid; Invasion; Migration

Core Tip: This systematic review highlights human stem cell-derived neural organoids as promising three-dimensional models for investigating glioblastoma stem cell invasion/migration. Despite considerable methodological heterogeneity, the studies demonstrate the potential of these models to replicate key aspects of the tumor microenvironment, assess therapeutic responses, and support personalized medicine approaches. However, the lack of standardized protocols and evaluation methods poses a challenge to reproducibility and broader translational use. Standardizing methodologies will be key to advancing these models and establishing their value in translational glioblastoma research and the development of targeted therapies.

INTRODUCTION

Glioblastoma multiforme (GBM) is the most common and lethal primary malignant tumor of the central nervous system in adults, accounting for approximately 45.2% of malignant intracranial tumors and 57.3% of all gliomas[1,2]. Globally, the incidence is estimated at 3.19-3.23 cases per 100000 people per year, with a higher prevalence in men (male-to-female ratio: 1.6:1 to 2.6:1) and a predilection for individuals over 60 years of age, who represent up to 83% of cases[2,3]. GBM is characterized by rapid growth, diffuse invasion, and intrinsic resistance to conventional treatments, leading to a median survival of 14.6 months, with fewer than 7.2% of patients surviving beyond five years[4,5].

One of the main challenges in treating glioblastoma is the presence of glioblastoma stem cells (GSCs), a subpopulation capable of self-renewal, differentiation and phenotypic plasticity. These properties enable resistance to apoptosis, immune evasion, and the ability to regenerate the tumor after therapy[1,6]. GSCs drive tumor recurrence and treatment failure, including surgery, radiotherapy, and chemotherapy[7,8], and are further protected by the blood-brain barrier and efflux transporters that limit the effectiveness of systemic therapies[7].

The brain microenvironment plays a decisive role in regulating GSC behavior, particularly their invasion/migration, and differentiation. GSCs interact with astrocytes, neurons, microglia, and the extracellular matrix (ECM), and have been shown to form functional synapses with glutamatergic neurons, which promotes tumor progression[1,9]. Moreover, GSCs use interconnected microtubule networks for cell-cell communication and dispersion, enhancing diffuse infiltration and therapeutic resistance[6,10]. Therefore, understanding the mechanisms of GSC invasion requires models that closely mimic the structural and functional complexity of the human brain.

Traditional models based on two-dimensional (2D) cultures, while simple and experimentally accessible, fail to reproduce the three-dimensional (3D) interactions and spatial complexity of the tumor microenvironment[11-14]. On the other hand, in vivo models, such as orthotopic murine xenografts, offer greater anatomical realism but are limited by ethical concerns, high costs, species-specific differences and low scalability[15,16].

In this context, 3D models derived from human stem cells, known as cerebral organoids, brain organoids or, more broadly, neural organoids (NOs), have emerged as powerful platforms for modeling neurodevelopment and, more recently, neuro-oncological processes[17,18]. NOs are self-organizing structures derived from human embryonic stem cells (hESCs) or induced pluripotent stem cells that mimic key aspects of human brain organization and development. They can be generated using guided differentiation protocols, where defined morphogens direct cells toward specific brain regions (e.g., cortex, thalamus), or through unguided protocols, in which cells spontaneously self-organize into heterogeneous brain-like tissues without predefined external cues[19-21].

Transcriptomic and histological analyses have shown that NOs faithfully recapitulate features of the developing human brain, including neuroepithelial proliferation, ventricular zones, mature neurons, astrocytes and functionally active microglia[17,22]. Incorporation of GSCs into these models has led to the development of systems such as glioma-cerebral organoid (GLICO), which replicate key aspects of tumor invasion, GSC proliferation, functional interaction with host neurons, and the formation of tumor-derived microtubule networks[23-25]. Although NOs have certain limitations such as not fully replicating features of the aged brain where glioblastoma most commonly arises, and lacking vascularization, a mature ECM, and key components of the tumor microenvironment, including the immune response[26], the value of using NOs to study GBM invasion/migration lies in their ability to provide a humanized 3D microenvironment[27]. Therefore, this systematic review aims to compile and critically evaluate the scientific literature on human stem cell-derived NO models used to study glioblastoma invasion/migration. It examines main models, culture methods and conditions, cell labeling strategies, NOs differentiation protocols, and the broader implications for personalized therapies and preclinical research.

MATERIALS AND METHODS

Search strategy

The articles included in this review were identified through searches conducted in the PubMed, Scopus, and Web of Science databases, following the PRISMA guidelines[28]. The search strategy prioritized the most recent and relevant literature, especially targeting studies published between March 2019 and March 2025. To align with the focus of this review, the following keywords were used: ((Spheroid) AND (Glioblastoma) AND (Stem Cell)), combined with Boolean operators and controlled vocabulary (DeCS/MeSH) on PubMed: (((spheroid[Title/Abstract]) OR (organoid[Title/Abstract])) AND (glioblastoma[Title/Abstract])) AND (“stem cell”[Title/Abstract]), on Scopus: (TITLE-ABS-KEY (organoid) OR TITLE-ABS-KEY (spheroid) AND TITLE-ABS-KEY (glioblastoma) AND TITLE-ABS-KEY (“stem cell”)), and on Web of Science: (((AB = (spheroid)) OR AB = (organoid)) AND AB = (glioblastoma)) AND AB = (“stem cell”).

Inclusion criteria

Only original articles published in English between March 2019 and March 2025 were included. The selected articles had to meet the following criteria: (1) Development and use of NOs models, associated with the culture and application of tumor spheroids or GBM cells; and (2) Assessment of invasion/migration using 3D co-culture models. These inclusion criteria were structured according to the population, exposure, outcome (PEO framework): (1) Population (P) - human NOs derived from stem cells; (2) Exposure (E) - exposure to GSCs or GBM tumor spheroids; and (3) Outcome (O) - evidence of tumor cell invasion/migration evaluated by histological analysis, immunofluorescence, live imaging, transcriptomics, or equivalent techniques, to investigating the role of NOs in tumor cell invasion/migration within 3D co-culture models of GBM.

Exclusion criteria

Articles were excluded in this review based on the following criteria: (1) Review articles; (2) Book chapters; (3) Editorials; (4) Short communications; (5) Conference abstracts or proceedings; (6) Letters to the editor; (7) Errata; (8) Articles not written in English; (9) Book series; (10) Duplicate entries across databases; (11) Studies that did not involve NOs; (12) Studies that assessed invasion using 2D culture systems, wound-scratch assays, Boyden chamber, ECM-based 3D migration assays, transwell inserts; and (13) Studies that did not involve the use of stem cells.

Data extraction

The selected articles were analyzed based on four key topics, each summarized in a dedicated table addressing the following aspects: (1) Characteristics and methodology for the formation of GBM models (including details on spheroid or cell-based approaches); (2) Characteristics and methodology for the formation of NO models; (3) Protocols of NO differentiation; and (4) Analysis of glioblastoma invasion/migration into NOs.

Data compilation and review

In this systematic review, a pre-selection of titles was performed by the authors Alves AH and Valle NME from the defined search strategies. The ten authors (Alves AH, Valle NME, Yokota-Moreno BY, Galanciak MCS, Silva KF, Mamani JB, Sertié AL, Oliveira FA, Nucci MP, Gamarra LF), in pairs, independently and randomly reviewed and analyzed the eligibility of the articles according to the selection criteria mentioned above. In case of discrepancy in study selection between two authors, the criteria were discussed with a third reviewer and resolved.

Alves ADH, Ennes do Valle NM, Galanciak MCDS, and de Oliveira FA researched for the characteristics and methodology for the formation of glioblastoma tumor model; Alves ADH, Ennes do Valle NM, Yokota-Moreno BY, and Felix da Silva K researched for characteristics of NO models; Alves ADH, Ennes do Valle NM, Yokota-Moreno BY, and Sertie AL researched for NOs differentiation protocols; Alves ADH, Ennes do Valle NM, Mamani JB, Nucci MP, and Gamarra LF researched for the analysis of the invasion of glioblastoma model into NOs. Analyses of data extracted from tables and flowcharts were performed by full peer consensus, respecting the above distribution. In this review, all authors wrote the entire text.

Risk of bias assessment

To minimize selection bias, the article screening and selection was carried out independently by two reviewers working in parallel. In cases of disagreement regarding the inclusion or exclusion of a study, a third independent reviewer was consulted to provide a final decision. The data extracted from the selected studies were compiled into structured tables and categorized according to predefined analytical groups. Each dataset was subsequently reviewed and validated by a second group of authors to ensure accuracy and consistency. The final decision regarding the inclusion of studies in this systematic review was made by consensus among all reviewers. This collaborative approach was designed to ensure methodological rigor and reduce the risk of bias in both the selection and analysis processes.

Methodological quality assessment

The methodological quality of the included studies was assessed using the Risk of Bias in Non-Randomized Studies of Interventions (ROBINS-I) tool, a widely recognized framework for evaluating the risk of bias in non-randomized experimental studies. The ROBINS-I tool evaluates seven key domains: Confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of the reported results[29]. Each domain is rated according to standardized criteria and classified as “Low”, “Moderate”, “Serious”, “Critical”, or “No information”. Ratings are determined based on the degree to which the study design and execution minimized the risk of bias within each domain[30]. The ROBINS-I assessment was independently conducted by three reviewers (Alves ADH, Ennes do Valle NM, and Mamani JB), all with experience in preclinical neurobiological research. To evaluate inter-rater reliability, the Fleiss’ kappa coefficient was calculated, which quantifies the level of agreement among more than two raters, ranging from 0 (no agreement) to 1 (perfect agreement)[31]. The calculations were performed using JASP software (v0.19.3; https://jasp-stats.org/; accessed on 25 May 2025). Visual representations of the risk-of-bias assessments were generated using the robvis R package and Shiny web application[32].

Statistical analysis

The variables presented in the summary tables and flowcharts were analyzed descriptively and presented as percentages to illustrate the distribution and key characteristics of the studies included in this systematic review. The most relevant and frequently reported findings were emphasized, while notable exceptions were also highlighted where applicable. In instances where substantial variability existed across studies, value ranges were reported to reflect methodological diversity and differences in experimental outcomes. To enhance the interpretability of the data, the results were organized into thematic categories, allowing for a structured and in-depth analysis of the reviewed literature.

RESULTS

Overview of the study selection procedure

The flowchart in Figure 1 summarizes the process of study identification, screening, eligibility assessment, and final inclusion in this systematic review, based on the predefined selection strategy and keywords. A total of 419 articles were initially identified through database searches, with 48 records retrieved from PubMed, 329 from Scopus, and 42 from Web of Science. During the initial screening phase, 10 review articles were excluded from the PubMed results. From the Scopus database, 141 records were excluded for the following reasons: 76 were review articles, 31 were duplicate entries, 8 were book chapters, 8 belonged to book series, 6 were editorials, 4 were not written in English, 3 were notes, 2 were conference papers, 1 was a conference proceeding, 1 was a letter to the editor, and 1 was an erratum. From Web of Science database, 42 were excluded for duplicate entries. Following this stage, 226 articles - 38 PubMed articles, 188 Scopus articles, and 0 Web of Science articles - were retained and subjected to full-text eligibility assessment.

Figure 1 Flowchart of the systematic review process based on PRISMA guidelines. It details the number of records identified, screened, and excluded, along with the main reasons for exclusion at each stage. A total of 10 studies met all criteria and were included in the final review.

During the full-text evaluation, 36 articles from PubMed were excluded for the following reasons: 19 articles did not include NOs, 14 did not assess invasion using 3D models, and 3 employed models that did not involve stem cells. Similarly, 180 records from Scopus were excluded based on the flowing criteria: 71 articles lacked NOs, 96 studies did not evaluate invasion/migration in 3D systems, and 13 did not utilize stem cell-based models. After applying all inclusion and exclusion criteria, a total of 10 original studies[14,23-25,33-38] were included in the final synthesis of this systematic review (Figure 1).

Methodological quality assessment

The risk of bias across the seven domains defined by the ROBINS-I was independently assessed for each included study by three reviewers. Overall, the majority of studies were judged to have low risk of bias in most domains, particularly in areas related to outcome measurement, missing data, and deviations from intended interventions. However, notable concerns emerged in the domains related to confounding (D1) and participant selection (D2), which were the most frequently rated as having moderate or serious risk of bias, reflecting common challenges in controlling for baseline differences and ensuring appropriate inclusion criteria in non-randomized designs. These patterns are visually summarized in Figure 2, where panel (A) illustrates the domain-level ratings for each study, and panel (B) shows the aggregated distribution of risk levels across all studies and domains, emphasizing the concentration of higher-risk judgments in D1 and D2.

Figure 2 Risk of bias assessment of the included studies using the Risk of Bias in Non-Randomized Studies of Intervention-I. A: Visual representation of the domain-level risk of bias judgments (D1 to D7) for each study. Each cell indicates the assigned risk level based on consensus among reviewers: Green (+) for low risk, yellow (–) for moderate risk, and red (×) for serious risk. The “Overall” column reflects the overall risk of bias rating across all seven domains; B: Summary plot showing the proportion of studies rated as “Low”, “Moderate”, or “Serious” risk of bias within each domain. Higher proportions of moderate or serious risk were observed in domains related to confounding (D1) and selection of participants (D2), while other domains predominantly showed low risk of bias.

To assess the reliability of these judgments, Fleiss’ kappa was calculated to determine inter-rater agreement across the three reviewers. The overall agreement was moderate [κ = 0.583, standard error = 0.152; 95% confidence interval (CI): 0.286-0.881]. Agreement was substantial for studies rated as having low risk of bias (κ = 0.830; 95%CI: 0.472-1.188), moderate for those classified with moderate risk (κ = 0.550; 95%CI: 0.192-0.908), and poor for those with a serious risk of bias (κ = -0.071; 95%CI: -0.429 to -0.287), suggesting increased variability in the more critical ratings.

Characteristics and methodology for the formation of glioblastoma tumor model (spheroids or cells)

The selected studies were analyzed for methodological aspects related to the establishment of GBM tumor models. Key parameters assessed and summarized in Table 1 and Figure 3 include the model generation method, cell type and seeding density, the expression of fluorescent markers to label GBM cells, and culture conditions, such as medium composition and incubation time.

Figure 3 Characterization of experimental models used to investigate glioblastoma invasion in neural organoids. The figure presents the main methodological parameters reported in the included studies, including tumor model type (suspension cells or spheroids), cell type and source (including glioblastoma stem cells and established line), fluorescent markers used, stem cell types used for organoid generation (human induced pluripotent stem cells, and human embryonic stem cells) and their sources. It also shows the number of tumor cells used in invasion/migration assays and the age of the organoids at the time of analysis. GSCs: Glioblastoma stem cells; GFP: Green fluorescent protein; EGFP: Enhanced green fluorescent protein; H2B-GFP: Histone-fused construct green fluorescent protein; hiPSCs: Human induced pluripotent stem cells; hESCs: Human embryonic stem cells; NR: Not reported.

Table 1 Characteristics and methodology for the formation of glioblastoma tumor model (spheroids or cells).

Ref.	Tumor model (method)	Cell number	Cell type	Cell/stem cell	Fluorescent expressed	Culture medium	Supplements	Culture time (days)
Van De Looverbosch et al[33], 2025	Suspension cells	1000 and 2000	Patient-derived	GSCs (LBT037-EGFP)	EGFP	Neurocult medium	bFGF, EGF, heparin, and anti-anti	NR
Ferreira et al[34], 2024	Suspension cells	100000	Lineage	U343MG	GFP	DMEM	Anti-anti and FBS	NR
				LN18
Pedrosa et al[25], 2023	Spheres (grow up as tumor spheres in non-laminin coated plate)	NR	Patient-derived	Proneural (GIC7)	GFP	DMEM/F12	EGF, bFGF, glucose, N2, glutamine, BSA, and HEPES	NR
				Mesenchymal (PG88)
Fedorova et al[35], 2023	Spheres (poly-HEMA-treated ULA V-bottom 96-well plate + centrifuged (200 g/2 minute)	2000	Lineage	U87MG	GFP and tdTomato	DMEM/F12	Glutamax, NEAA, PS, and FBS	1
Bassot et al[36], 2023	Suspension cells	NR	Patient-derived	Classical GSC (Ge904)	Non-labeled	DMEM-high	Glucose, Glutamax, PS, and FBS	NR
Goranci-Buzhala et al[37], 2021	Suspension cells	1000	Patient-derived	GSCs (U3047MG (OPC-like)	mCherry	Neurocult NS-A	EGF, bFGF, N2, L-glutamine, B27 without vitamin A, heparin, and BSA	NR
				GSCs (U3024MG (MES-like)
Azzarelli et al[14], 2021	Suspension cells	10000 or 50000	Patient-derived	GSCs (GCGR-E27 and GCGR-E35)	H2B-GFP	D8437	EGF, bFGF, N2, Glutamax, B27, PS, and laminin	NR
Krieger et al[38], 2020	Suspension cells	1000	Patient-derived	4 patient-derived cell lines	GFP	Neurobasal medium	EGF, bFGF, L-glutamine, B27, and heparin	7-21
Goranci-Buzhala et al[23], 2020	Spheres (ULA U-bottom 96-well plate)	1000	Patient-derived	GSCs primary lines (U3047MG, U3024MG, 450)	mCherry	Neurocult NS-A	EGF, bFGF, N2, L-glutamin, B27 without vitamin A, heparin, and BSA	2
				GSCs recurrent lines (275-BIS)	GFP
Linkous et al[24], 2019	Suspension cells	10000	Patient-derived	GSCs	GFP	Neurobasal medium	bFGF, sodium, N2, B27, L-glutamine, and heparin	NR
					RFP-827

GSCs: Glioblastoma stem cells; EGFP: Enhanced green fluorescent protein; bFGF: Basal fibroblast growth fator; EGF: Epidermal growth factor; NR: Not reported; GFP: Green fluorescent protein; DMEM: Dulbecco’s Modified Eagle Medium; FBS: Fetal bovine serum; BSA: Bovine serum albumin; HEPES: 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid; Poly-HEMA: Poly-hydroxyethyl methacryla; ULA: Ultra-low attachment; NEAA: Non-essential amino acids; PS: Penicillin/streptomycin; OPC-like: Oligodendrocyte-progenitor-like; MES-like: Mesenchymal-like; H2B-GFP: Histone-fused construct green fluorescent protein; RFP: Red fluorescent protein; Anti-Anti: Antibiotic/antimycotic.

With respect to tumor model generation, 70% of the selected studies reported using suspension-based GBM models[14,24,33,34,36-38], which involve culturing dissociated tumor cells in a suspension environment, while the remaining 30% employed spheroid-based models[23,25,35], where tumor cells aggregate into compact, spherical structures. Among the spheroid models, one study used ultra-low attachment (ULA) U-bottom 96-well plates[23], another applied a poly-hydroxyethyl methacryla-treated ULA V-bottom plate followed by centrifugation[35], and a third utilized non-laminin-coated plates[25].

In terms of cell seeding density, suspension-based systems typically employed between 1000 and 10000 GBM cells per culture. Notable exceptions include studies by Azzarelli et al[14] and Ferreira et al[34] which used 50000 and 100000 cells, respectively. For spheroid-based models, the number of cells per spheroid ranged from 200 to 2000[23,25,35], reflecting methodological variability in model scale and density across studies.

Regarding the cellular origin of the GBM models, 80% of the studies used patient-derived GBM cells[14,23-25,33,36-38], while the remaining 20% used established cell lines, including U343MG and LN18[34], as well as U87MG[35]. Among the eight studies that used patient-derived GBM cells[14,23-25,33,36-38], six exclusively selected GSCs for tumor model development[14,23,24,33,36,37]. The remaining two studies employed differentiated tumor cells[25] and non-specified patient-derived GBM cells[38].

Most of the selected studies (90%) employed a range of fluorescent proteins to label GBM cells, classified by color emission and functional characteristics[14,23-25,33-35,37,38]. These markers enabled real-time tracking during invasion assays and enhanced the visualization of tumor dynamics in co-culture systems. Six studies used green fluorescent protein (GFP) in its conventional[23-25,34,35,38]. Additionally, Van De Looverbosch et al[33] and Azzarelli et al[14] employed GFP variants, specifically enhanced GFP and a histone-fused construct GFP, respectively. Other fluorescent markers were also employed across studies. Linkous et al[24] used red fluorescent protein (RFP), while Fedorova et al[35] incorporated tdTomato, a bright RFP, into their labeling strategy. Furthermore, two studies reported the use of mCherry, an RFP with distinct spectral properties, to label GBM cells[23,37]. Only one study did not employ any fluorescent labeling for GBM cell tracking[36].

Regarding the culture media used for GBM cells, 40% of the studies employed Dulbecco’s Modified Eagle Medium (DMEM)-based formulations, including DMEM/F12[25,35], DMEM-high glucose[36], and standard DMEM[34]. Neurobasal^™ medium was used in 20% of the studies[24,38], while NeuroCult^™ was used in another 30%, in its standard[33] and NS-A[23,37] formulations. Only one study, conducted by Azzarelli et al[14], reported the use of D8437 medium. An analysis of the supplements used across the 10 studies revealed a widespread inclusion of growth factors, reported in 70% of the articles[14,23-25,33,37,38], particularly epidermal growth factor (EGF) and basic fibroblast growth factor (bFGF). Metabolic support supplements were included in 80% of the studies[14,23-25,35-38], comprising Glutamax, L-glutamine, N2, B27, non-essential amino acids (NEAA), glucose, and sodium - all critical for maintaining cellular growth, survival and nutritional homeostasis. With regard to antimicrobial agents, 50% of studies used antibiotics/antimycotics[14,33-36], such as antibiotic/antimycotic and penicillin/streptomycin. ECM-associated components, including laminin and heparin, were utilized in 60% of the studies analyzed[14,23,24,33,37,38], reflecting their role in mimicking the native tumor microenvironment. Furthermore, serum-derived supplements, such as fetal bovine serum[34-36] and bovine serum albumin[23,25,37], were incorporated in an equal proportion of studies.

Among the studies that employed GBM spheroids as a model[23,25,35], the culture medium used for spheroid formation was the same as that applied in the corresponding 2D cell cultures. Only three studies (30%) reported the duration of GBM cell culture prior to co-culture with NOs, with timeframes ranging from 1 to 21 days, specially 1 day[35], 2 days[23], and 7 to 21 days[38]. Notably, in all studies using GBM spheroids, the use of ECM components for spheroid formation was not reported.

Characteristics of NOs models

The selected studies used NOs as 3D host tissues to assess GBM cell infiltration and tumor-host interactions. Key characteristics of NOs, including source of stem cells, seeding density, culture duration, organoid size, and phenotypic characterization markers are summarized in Table 2 and Figure 3.

Table 2 Characteristics of neural organoid models.

Ref.	Cell/stem cell	iPSCs source	Cell number	NOs age (days)	Assessment size	Characterization1
Van De Looverbosch et al[33]	hiPSCs (iPSC0028)	Epithelium	10000	30	1077 ± 272 μm	NR
Ferreira et al[34]	hiPSCs (F9048)	Skin fibroblasts	10000	40	NR	SOX2, TUJ1, DCX
Pedrosa et al[25]	hiPSCs (BJiPSC-SV4F-9)	Skin fibroblasts	NR	41	NR	SOX2, TUJ1, GFAP, O4
Fedorova et al[35]	hiPSCs (MUNIi008-A)	Skin fibroblasts	2000-3000	55	NR	PAX6, TUJ1, DCX, MAP2, BRN2, SYN1
	hiPSCs (MUNIi009-A)
	hiPSCs (MUNIi010-A)
Bassot et al[36]	hiPSCs + hESCs (HS420)	Fetal skin fibroblasts + Blastocyst	1000	40	NR	TUJ1, MAP2, NeuN, GFAP, S100β, OLIG2
Goranci-Buzhala et al[37]	hiPSCs (IMR90)	Fetal lung fibroblast	35000	10	NR	NR
Azzarelli et al[14]	hiPSCs (IMR90)	Fetal lung fibroblast	9000	42	NR	SOX2, N-cadherin, PAX6, TUJ1, TBR1
Krieger et al[38]	hiPSCs (409b2)	Skin fibroblasts	1000	24	500-900 μm	PAX6, TUJ1
Goranci-Buzhala et al[23]	hiPSC (IMR90) + GSCs	Fetal lung fibroblast	35000 (hiPSC) + 1000 (GSC)	20, 40 and 60	500-700 μm	MAP2, SYN1
			35000 (hiPSC) + 5000 (GSC)
Linkous et al[24]	hESCs (WA01 and WA09)	Blastocyst	9000	NR	NR	Nestin, musashi-1, SOX2, PAX6 and TBR2
	hiPSCs (H6)	NR

¹Were considered only positive markers.

iPSCs: Induced pluripotent stem cells; NOs: Neural organoids; hiPSCs: Human induced pluripotent stem cells; NR: No reported; SOX2: Sex determining region box 2; TUJ1: Beta-tubulin III; DCX: Doublecortin; GFAP: Glial fibrillary acidic protein; PAX6: Paired box protein 6; BRN2: BEARSKIN2; SNY1: Synapsin1; hESCs: Human embryonic stem cells; MAP2: Microtubule-associated protein 2; NeuN: Neuron-specific nuclear; OLIG2: Oligodendrocyte transcription factor 2; S100β: Calcium-binding protein β; GSCs: Glioblastoma stem cells.

Data analysis reveals that most studies focused on the use of human induced pluripotent stem cells (hiPSCs), with eight employing hiPSCs exclusively[14,24,25,33-35,37,38]. An exception was the study by Linkous et al[24], which used both hiPSCs (line H6) and hESCs (lines WA01 and WA09) for organoid generation. Only one study reported the co-culture of hiPSCs with GSCs[23]. Regarding the stem cell source, five studies used skin-derived fibroblasts for hiPSC generation[25,34-36,38], with Bassot et al[36] specifically indicating fetal tissue. Three studies reported using fetal lung fibroblasts[14,23,37], while Van De Looverbosch et al[33] identified the hiPSC source as epithelial tissue. One study did not specify the source of the hiPSCs used[24]. For hESCs-derived models, both studies reported the blastocyst was cited as the original source[24,36].

The initial number of pluripotent cells used for NO formation ranged from 1000[36,38] to 35000 cells[23,37], which approximately 10000 cells emerging as the most frequently employed quantity - reported in 40% of the studies[14,24,33,34]. Fedorova et al[35] used an intermediate range of 2000-3000 cells, while Pedrosa et al[25] did not specify the seeding density. Only three studies reported the final size of the NOs used for invasion analysis, with diameters ranging from 500-700 μm[23], 500-900 μm[38], and 1077 ± 272 μm[33]. The culture duration (NO age) ranged from 10 to 60 days, with a predominant culture time of 40 ± 2 days, reported in five studies[14,23,25,34,36]. One study used NOs cultured for 20, 40, and 60 days[23]. Less common culture times included 10 days[37], 24 days[38], 30 days[33], and 55 days[35]. The study by Linkous et al[24] did not specify the age of the NOs used.

Neural differentiation was primarily assessed through immunolabeling, with beta-tubulin III being the most frequently reported neuronal marker, used in 60% of the studies[14,25,34-36,38]. Other, more specific, neuronal markers were used less frequently as microtubule-associated protein 2 (dendritic marker)[23,35,36], doublecortin (immature neuron marker)[23,35,36], neuron-specific nuclear (mature neuron marker)[36], as well as BEARSKIN2[35], TBR1[14,24,25,34] and TBR2[24] (regional identity markers). Among markers for neural progenitor cells (NPCs) and pluripotency, sex determining region box 2[14,24,25,34] and paired box protein 6[14,24,35,38] were the most commonly used. Other NPC markers such as nestin[24] and musashi-1[24] appeared in a smaller subset of studies. For glial lineage characterization, GFAP[25,37] was the most frequently used astrocyte marker, followed by S100β[36], and the oligodendrocyte markers O4[25] and oligodendrocyte transcription factor 2[36], that appeared less often. N-cadherin[14], a key cell adhesion molecule, was analyzed in one study, and the synaptic marker synapsin1 was analyzed in two studies[23,35]. Regarding in vitro tracking of NOs, only the study by Goranci-Buzhala et al[37] reported the use of cells transfected with the fluorescent marker tubulin-GFP.

NOs differentiation protocols

NOs generation generally follows three main phases. It begins with the formation of embryoid bodies (EBs), which are 3D structures derived from pluripotent stem cells that recapitulate aspects of early embryonic development. After that or concurrently, neural induction is initiated using specific medium to drive stem cells differentiation into neuroepithelium, neural stem cells and NPCs. Finally, during the neural differentiation/maturation phase, a different medium promotes the development of these progenitors into neurons and glial cells. In guided protocols, additional signaling molecules are used to direct differentiation toward specific brain regions, whereas unguided protocols allow for more spontaneous, self-organized development without regional specification.

Our analysis of the included studies revealed that 50% utilized unguided protocols for NO generation[14,24,34,35,38], while 40% employed guided approaches to direct differentiation specially towards cortical neural tissue[23,25,36,37]. Due to the use of a non-standard differentiation approach, one study (10%) could not be definitively classified as guided or unguided[33]. In addition to the differentiation strategy, we examined critical experimental parameters, including NO differentiation phases, plate configuration, culture media composition, supplementation, growth factors, ECM utilization, and the duration of each differentiation phase, as depicted in Table 3 and Figure 4A.

Figure 4 Overview of neural organoid generation protocols, glioblastoma multiforme co-culture strategies, and analytical methodologies. A: Top: Schematic representation of the main phases involved in the generation of neural organoids (NOs), including: (1) Embryoid body formation; (2) Neural induction; (3) Simultaneous embryoid body formation and neural induction; (4) Embedding in extracellular matrix; (5) Neural differentiation and maturation; and (6) Co-culture with 2D glioblastoma multiforme (GBM) cells or three-dimensional GBM spheroids. Most of the factors applied during these steps were specific to guided protocols designed to direct NOs toward a cortical identity, with the exception of the ROCK inhibitor, which was universally used in the early stages to enhance cell viability. Bottom: Temporal timeline (in days) of each phase as applied in the studies included in this review; B: Summary of methodologies used to assess GBM cell migration and invasion in NO models. Migration and invasion can be visualized using fluorescence or confocal microscopy, either by: (1) Tracking fluorescently labeled GBM cells on the surface of whole NOs; (2) Visualizing internal invasion using optical clearing techniques in whole NOs; and (3) Sectioning organoids into cryoslides for detailed imaging of GBM cells invasion. Alternatively, GBM cell invasion can be quantified by dissociating co-cultured NOs into single cells and analyzing them via flow cytometry, enabling the measurement of the proportion of GBM cells that infiltrated the organoids. ROCKi: ROCK inhibitor; bFGF: Basic fibroblasts growth factor; DM: Dorsomorphin; TGF: Transforming growth factor; BMPi: Bone morphogenetic protein inhibitor; EGF: Endothelial growth factor; BDNF: Brain-derived neurotrophic factor; GDNF: Glial cell line-derived neurotrophic factor; EB: Embryoid body; ECM: Extracellular matrix; EGFP: Enhanced green fluorescent protein; GFP: Green fluorescent protein; H2B-GFP: Histone-fused construct green fluorescent protein; RFP: Red fluorescent protein; NO: Neural organoid.

Table 3 Neural organoids differentiation protocols.

Ref.	Protocol	NOs phases	Plate	Medium	Supplements	Factors	ECM used	Time of each phase
Van De Looverbosch et al[33]	N/A	2D hiPSCs neural induction	N/A	Neurobasal + DMEM/F12 + Glutamax	Glutamax, MEM-NEAA, sodium pyruvate, 2-ME, human insulin, N2s, B27s, PS	LDN-193189 and SB-431542	-	11 days
		2D NPCs maturation	N/A	Neurobasal + DMEM/F12 + Glutamax	Glutamax, MEM-NEAA, sodium pyruvate, 2-ME, human insulin, N2s, B27s, PS	-	-	19 days
		Organoid production	U-bottom 96-well1	-	-	-	-	7 days
Ferreira et al[34]	Unguided	EB formation	ULA U-bottom 96-well	mTeSR 1	Normocin	ROCKi	-	Not clear
		Neural induction	ULA U-bottom 96-well	DMEM/F12	Knockout serum replacement, MEM-NEAA, Glutamax, 2-ME	ROCKi and bFGF	-	Until EBs reached 400-600 μm
		Neural induction	ULA 24-well	DMEM/F12	N2s, MEM-NEAA, heparin	-	-	Until neuroepithelium cues appeared
		ECM embedding	ULA 6-well	Neurobasal + DMEM/F12	N2s, B27s without vitamin A, MEM-NEAA, Glutamax, 2-ME, human insulin, Normocin	-	Matrigel	4 days
		Neural differentiation/maturation	ULA 6-well on orbital shaker	Neurobasal + DMEM/F12	N2s, B27s with vitamin A, MEM-NEAA, Glutamax, 2-ME, human insulin, Normocin	-	-	Up to use in co-cultures (> 40 days)
Pedrosa et al[25]	Guided	EB formation/neural induction	ULA V-shaped 96-well	DMEM/F12	KnockOut serum, MEM-NEAA, Glutamax, 2-ME, PS	ROCKi, DM and, SB-431542	-	7 days
		Neural differentiation	96-well	Neurobasal-A	B27 supplement without vitamin A, Glutamax, and PS	bFGF and EGF	-	14 days
		Neural differentiation	ULA 6-well	Neurobasal-A	B27s without vitamin A, Glutamax, and PS	bFGF, EGF, BDNF and, NT-3	-	21 days
		Neural maturation	ULA 6-well	Neurobasal-A	B27s without vitamin A, Glutamax, and PS	-	-	Up to use in co-cultures (> 42 days)
Fedorova et al[35]	Unguided	EB formation	V-bottom 96-well2	mTeSR 1	-	ROCKi	-	Until EB were at least 400 μm in diameter
		Neural induction	ULA 24-well	DMEM/F12	N2s, Glutamax, MEM-NEAA and, heparin	-	-	6 days
		ECM embedding	6 cm dish	Neurobasal + DMEM/F12	Glutamax, MEM-NEAA, N2s, human insulin, 2-ME, B27s without vitamin A and, PS	-	Geltrex	4 days
		Neural differentiation/maturation	Spinning bioreactor	Neurobasal + DMEM/F12	Glutamax, MEM-NEAA, N2s, human insulin, 2-ME, B27s with vitamin A and, PS	-	-	Up to use in co-cultures
Bassot et al[36]	Guided	EB formation	Microwell plate	Serum-free medium	-	ROCKi	-	12-36 hours
		Neural induction	6-well on orbital shaker	Neurobasal + DMEM/F12 + Glutamax	B27s and MEM-NEAA	TGFβ/activin/Nodali and BMPi	-	4 days
		Neural differentiation	6-well on orbital shaker	Neurobasal + DMEM/F12 + Glutamax	-	EGF, bFGF, BMPi, GDNF, BDNF and iγ-secretase	-	17 days
		Neural maturation	PTFE membrane 6-well without agitation	Neurobasal + DMEM/F12 + Glutamax	-	GDNF, BDNF and iγ-secretase	-	Up to use in co-cultures
Goranci-Buzhala et al[37]	Guided	EB formation/neural induction	ULA 96-well	NIM	-	ROCKi		5 days
		ECM embedding	NR	Neurobasal + DMEM/F12	N2s, B27s without vitamin A, 2-ME, human insulin, L-glutamin and, MEM-NEAA	-	Matrigel	4 days
		Neural differentiation/maturation	Spinner flask	Neurobasal + DMEM/F12	N2s, B27s without vitamin A, 2-ME, human insulin, L-glutamin, MEM-NEAA	DM; SB-431542	-	Up to use in co-cultures (> 10 days)
Azzarelli et al[14]	Unguided	EB formation	ULA 96-well	EB-media (basal 1 plus supplement A)	-	ROCKi		5 days
		Neural induction	ULA 24-well	NIM (basal 1 plus supplement B)	-	-	-	2 days
		ECM embedding	6 cm dish	Expansion media (basal 2 plus supplement C and D)	-	-	Matrigel	3 days
		Neural differentiation/maturation	6 cm dishes on orbital shaker	Maturation medium (basal 2 plus supplement E)	Matrigel dissolved in the medium	-	-	Up to use in co-cultures (> 10 days)
Krieger et al[38]	Unguided	EB formation/neural induction	AggreWell	NIM	-	ROCKi	-	5 days
		ECM embedding	NR	Neurobasal + DMEM/F12 + Glutamax	1:1 mixture of (N2s, human insulin, L-glutamine, MEM-NEAA, 2-ME): B27s, L-glutamine and, PS	-	Matrigel	Not clear
		Neural differentiation/maturation	NR	Neurobasal + DMEM/F12 + Glutamax	1:1 mixture of (N2s, human insulin, L-glutamine, MEM-NEAA, 2-ME): B27s, L-glutamine and, PS	-	-	Up to use in co-cultures
Goranci-Buzhala et al[23]	Guided	EB formation/neural induction	ULA 96-well	NIM	-	ROCKi	-	5 days
		ECM embedding	NR	Neurobasal + DMEM/F12	N2s, B27s without vitamin A, 2-ME, human insulin, L-glutamin, MEM-NEAA	-	Matrigel	4 days
		Neural differentiation/maturation	Spinner flask	Neurobasal + DMEM/F12	N2s, B27s without vitamin A, 2-ME, human insulin, L-glutamin, MEM-NEAA	DM and SB-431542	-	Up to use in co-cultures (> 10 days)
Linkous et al[24]	Unguided	EB formation/neural induction	LA 96-well	NIM	-	ROCKi	-	6-7 days
		Rosettes formation	Matrigel-coated 6-well	NIM	-	-	-	11-14 days
		ECM embedding	NR	Neurobasal + DMEM/F12	N2s, B27s without vitamin A, 2-ME, human insulin, Glutamax, MEM-NEAA	-	Matrigel	4 days
		Neural differentiation/maturation	Spinning bioreactor or orbital shaker	Neurobasal + DMEM/F12	N2s, B27s with vitamin A, 2-ME, human insulin, Glutamax and, MEM-NEAA	-	-	Up to use in co-cultures (> 25 days)

¹Treated anti-adherence solution.

²Poly-hydroxyethyl methacryla-treated non-adherence.

NOs: Neural organoids; ECM: Extra cellular matrix; 2D: Two-dimensional; hiPSCs: Human induced pluripotent stem cells; N/A: Not applicable; DMEM: Dulbecco’s Modified Eagle Medium; MEM-NEAA: Minimum essential media non-essential amino acids; 2-ME: 2-mercaptoetanol; N2s: N2 supplement; B27s: B27 supplement; PS: Penicillin/streptomycin; NPC: Neural progenitor cells; EB: Embryoid body; ULA: Ultra-low attachment; NR: No reported; LA: Low attachment; ROCKi: ROCK inhibitor; bFGF: Basal fibroblast growth factor; DM: Dorsomorphin; EGF: Epidermal growth factor; BDNF: Brain-derived neurotrophic factor; NT-3: Neurotrophin-3; TGFβ: Transforming growth factor β; Nodali: Nodal inhibitor; BMPi: Bone morphogenetic proteins inhibitor; GNDF: Glial derived neurotrophic factor; NIM: Neural induction medium.

In general, the analyzed studies initiated NOs generation with the formation of EBs, followed by neural induction, neurodifferentiation, and maturation[14,23-25,34-38]. In some studies, EB formation and neural induction occurred simultaneously[23-25,37,38], while in others, these steps were performed as distinct phases[14,34-36]. One study adopted an alternative approach, culturing hiPSCs and differentiating them into NPCs in a 2D monolayer before initiating organoid generation[33]. Regarding the duration of each phase, protocols varied across studies. While most studies defined phase transitions based on culture time[14,23-25,33-38], some incorporated additional criteria such as, organoid size[34,35] or the appearance of neuroepithelial structures[34].

Regarding the preparation of culture media, one study (10%) exclusively on commercial media throughout the differentiation process[14], five studies (50%) prepared all media formulations in-house[25,33-36], and the remaining four studies (40%) combined commercial media during the initial stages (EB formation and neural induction), followed by in-house formulations during the differentiation and maturation phases of the organoids[23,24,37,38]. Among the studies employing in-house media for sequential EB formation and neural induction, mTeSR1 was used for EB generation in two studies[34,35], while one employed an unspecified serum-free medium[36]. For the neural induction phase, DMEM/F12 used in two studies[34,35], and a 1:1 mixture of Neurobasal and DMEM/F12 supplemented with Glutamax[36]. The single study performing simultaneous EB formation and neural induction with in-house media used DMEM/F12[25]. During the neurodifferentiation and maturation phases, studies that prepared media in-house used either a 1:1 mixture of Neurobasal and DMEM/F12[23,24,34,35,37], a similar mixture supplemented with Glutamax[33,36,38], or Neurobasal A[25].

In terms of the supplements used, during the neuroinduction phase the most commonly reported were minimum essential media-NEAA[25,33-36], Glutamax[25,33-35] and 2-mercaptoethanol[25,33,34], followed by N2 supplement[33,35], B27 supplement[33,36], knockout serum replacement[25,34], and heparin[34,35]. Less frequently used supplement included human insulin solution[33] and sodium pyruvate[33]. Antibiotics, such as penicillin-streptomycin were used in two studies[25,33]. During the neurodifferentiation and maturation phases, most studies demonstrated a higher degree of consistency in medium composition, with only minor variations. Notably, Goranci-Buzhala et al[37] was the only study that did not add any supplements in the organoid culture medium. In contrast, other studies consistently included minimum essential media-NEAA, N2 supplement, insulin, and 2-mercaptoethanol[23,24,33-35,37,38]. Glutamax was also frequently included[24,25,33-35] and B27 supplement was included in all 9 studies, with some explicitly indicating the presence[24,33-35,38] or absence[23,25,37] of vitamin A. Additional supplements included L-glutamine[23,37,38] and sodium pyruvate[33]. Regarding antibiotics, penicillin-streptomycin was used in several studies[25,33,35,38], while Normocin was reported in one study[34].

Regarding differentiation factors, the ROCK inhibitor was employed in all studies during the EB formation phase across all 9 studies to improve survival[14,23-25,34-38]. For guided protocols, several signaling pathway inhibitors were included, particularly SMAD pathway inhibitors such as dorsomorphin and SB-431542[23,25,37], along with specific inhibitors targeting the transforming growth factor β/activin/nodal and bone morphogenetic protein pathways[36]. Commonly used growth factors included EGF, bFGF, brain-derived neurotrophic factor, and neurotrophin-3[25], as well as EGF, bFGF, bone morphogenetic protein inhibitor, glial cell line-derived neurotrophic factor, brain-derived neurotrophic factor, and γ-secretase inhibitor[36].

Finally, seven studies (70%) included an additional step in organoid generation by embedding the developing structures in ECM droplets, using either Matrigel[14,23,24,34,37,38] or Geltrex[35]. Notably, in three of these studies (30%), the B27 supplement without vitamin A was used during the ECM embedding phase and was later replaced with B27 containing vitamin A, during the differentiation and maturation stages[24,34,35]. This step contributed to the structural integrity of the developing organoids and may also influence their functional maturation.

Analysis of glioblastoma invasion into NOs

The invasion of the GBM models into the NOs was analyzed across the selected studies. Data were categorized according to co-culture methods, co-culture duration, techniques for invasion assessment, outcomes, and overall, the research objectives. A summary of these parameters is presented in Table 4 and Figure 4B.

Table 4 Analysis of the invasion of glioblastoma model into neural organoids.

Ref.	Co-culture method	Co-culture time (days)	Groups evaluated	Invasion analysis method	Invasion analysis results	General purpose
Van De Looverbosch et al[33]	GSCs cells were plated onto the NOs	13	1000 GSCs	Fluorescence images and computing mapping (with clearing)	For both samples, most isolated GSCs were found dispersed between 25 and 200 μm from their surface	To demonstrate the value of computing mapping based on deep learning for counting cells in spheroids, identifying differentially labeled subpopulations in spheroids, and mapping the invasion of GBM cells into NOs
			2000 GSCs
Ferreira et al[34]	GBM cells were plated onto the NOs ULA 6-well plates with shaker (65 rpm)	14	Mock group (7 and 14 days)	Confocal fluorescence images and flow cytometry	Significant reduction in tumor cell proportion on ZIKV group and all CNS tumor cell lines reached the inner regions of the NOs after two weeks	To establish a co-culture model of human NOs with cancer cells from various CNS tumors and utilize these assemblies to investigate the oncolytic effects of ZIKV
			ZIKV infection group (20000 PFU) (7 and 14 days)
Pedrosa et al[25]	GBM cells or spheroids were plated onto the NOs	41	2000 GBM cells	Confocal fluorescence images and phase-contrast images	GFP-GICs started to integrate 24 hours after co-culture began. By 15 days, GFP-GIC7 and GFP-PG88 attached to and entered the Nos. By 41 days, both had successfully invaded and multiplied inside the NOs	To evaluate 5-ALA-mediated PDT for GBM treatment, aiming to eliminate infiltrating tumor cells while preserving normal tissue using GBM-initiating cells (GIC7 and PG88) in NOs
			2 tumor spheres
Fedorova et al[35]	GBM spheroids were placed on top of NOs (inclined at 45°)	30, 60, and 90	Inclined plane	Confocal fluorescence images (with clearing)	GBM migration into NOs increases over time. Migration distance and cell number are significantly higher at 60 and 90 days, with the highest number of migrating cells at 60 days. Matrigel and Geltrex increased GBM cell migration to NOs compared to the system without ECM. Matrigel showed a higher number of cells distant from the GBM/NOs border compared to Geltrex	To propose a GLICO model to study GBM growth and migration in NOs, highlighting the impact of ECM proteins
	GBM spheroids were placed of NOs embedded in a droplet of Matrigel or Geltrex in a ULA dishes with orbital shaker (0.035 g)		Matrigel
			Geltrex
Bassot et al[36]	GSCs cells were plated onto the NOs	5	Control (miR-ctrl)	Immunofluorescence and confocal imagens	Observed invasion of the GSCs into the NOs in the miR-Ctrl condition and a stronger signal for Ki-67 in the invasive single cells distant from their primary site, indicating that miRNAs can penetrate the NOs and affect GBM invasive capacity	To identify miR-17-3p, miR-222, and miR-340 as key regulators of GBM aggressiveness. Their combined modulation inhibits tumor growth, induces cell death, and shows therapeutic potential in GBM models
			Transfected (miR-17-3p, miR-222, miR-340)
Goranci-Buzhala et al[37]	GSCs cells were plated onto the NOs in a ULA-Lumox dish	20	Nek2-KD protein expressed with DOX	Immunofluorescence and confocal imagens (with clearing)	Nek2-KD-expressing GSCs failed to enter brain organoids. Naive cells diffused and recapitulated the known characteristics of invading GSCs, such as establishing protrusion-like processes in the form of microtubes. Nek2-KD-expressing U3047MG cells exhibited impaired invasion and failed to grow in NOs	To explore how GSCs suppress ciliogenesis to maintain self-renewal and tumor progression. Restoring cilia formation induces GSC differentiation, reducing tumor infiltration. The study suggests cilium induction as a potential therapy for GBM
			Naíve (control)
Azzarelli et al[14]	GSCs cells were plated in ULA 96-well plate onto NOs 6 cm dishes	7	10000 cells	Confocal fluorescence images	On low density, the GSCs group shows some individual cells dispersed sparsely across NOs, the invasion by interconnected streams was more prominent in high-density GSCs group. The behavior of the two cell lines was comparable (genetic profile, classified as RTKI and present PDGFRα amplification)	Followed the behavior of two different cell lines in the GOC system and found that GSCs with similar genetic alterations exhibit comparable behaviors upon organoid engraftment
			50000 cells
Krieger et al[38]	GBM cells were plated onto the NOs in a GravityTRAP-ULA + centrifuged (100 g per 3 minutes)	2	Patient F2	Immunohistochemistry and confocal images (with clearing)	To similar NOs size, the fraction of NOs volume taken up by tumor cells was similar across the 4 patient-derived cell lines, with similar sizes. Tumor cells spread widely in all cases. Invasion depths exceeded 100 μm in the majority, with some cells detected at 300 μm from the surface. Cells from patients F6 and F9 were less invasive than cells from F2 and F3	To investigate GBM invasion in a human-derived model using NOs as a scaffold. This model provides a clinically relevant platform for studying GBM
			Patient F3
			Patient F6
			Patient F9
Goranci-Buzhala et al[23]	GSCs spheres were plated onto the NOs	3-10	20 days-old organoids	Fluorescence (with clearing) and time-lapse images	GSCs integrated faster into older organoids, suggesting that mature NOs create a favorable environment for GSC growth, likely driven by neuronal activity and factors like NLGN3. Additionally, differences in invasive patterns were shown between primary and recurrent GSCs	To establish three different methods to assess GSC invasion in NOs and develop imaging techniques to validate organoids as reliable models. The assays reveal GSC affinities for organoids at different stages and capture key invasion aspects
			40 days-old organoids
			60 days-old organoids
Linkous et al[24]	GSCs cells were plated in 24-well plate onto NOs with shaker	14	Control group	Fluorescence (with clearing), luciferase activity and histopathological	GSC-827 exhibited infiltrative edges with tumor cells invading normal NOs, while GSC-923 showed a diffuse invasion pattern, resembling GBM patient samples. Both cell types (827 and 923) formed a network of tumor microtubes that facilitated infiltration into NOs	To train GLICO models for high throughput drug screening
			TMZ group
			BCNU group
			Ionizing radiation (0, 5 and 10 Gy)

GSCs: Glioblastoma stem cells; NOs: Neural organoids; GBM: Glioblastoma; ULA: Ultra-low attachment; ZIKV: Zika virus; PFU: Plaque-forming units; CNS: Central nervous system; GFP-GICs: Glioblastoma cell line with green fluorescent protein expression; 5-ALA: 5-aminolevulinic acid; PDT: Photodynamic therapy; GLICO: Glioma-cerebral organoid; ECM: Extracellular matrix; Ki67: Proliferation marker; Nek2-KD: Nek-knockdown; DOX: Doxycycline; RTKI: Receptor tyrosine kinase I; PDGFRα: Platelet-derived growth factor receptor alpha; GOC: Glioblastoma-organoid co-cultures; NLGN3: Neuroligin-3; TMZ: Temozolomide; BCNU: Carmustine.

There was no single co-culture predominant co-culture method, reflecting the variability in experimental designs. Six studies utilized specialized culture plates, including ULA-well plates[14,34,35,37], GravityTRAP-ULA[38], and standard 24-well plates[24]. Four studies incorporated additional steps to enhance interaction between GBM cells and NOs: Three used orbital shaker[24,34,35], one employed centrifugation[38], and one tilted the culture plate at a 45°[35]. Notably, Fedorova et al[35] evaluated three distinct co-culture methods including conditions with and without ECM components such as Matrigel and Geltrex.

The duration of co-culture between tumor cells (either GSCs or GBM cells) and NOs varied substantially across the studies. Three studies (30%) employed a co-culture period of approximately 14 days[24,33,34], which was the most frequently reported duration. Shorter durations included 2 days[38], 5 days[36], and 7 days[14]. Goranci-Buzhala et al[23] adopted a flexible approach with a co-culture window ranging from 3 to 10 days. In contrast, extended co-culture durations were also reported, including 20 days[37], 41 days[25], and 30, 60, and up to 90 days in the most prolonged model[35].

Fluorescence-based imaging was the primary method for assessing tumor invasion, used in seven studies[14,23-25,33-35]. Four of these employed confocal microscopy[14,25,34,35], while two studies utilized immunofluorescence[36,37], one used immunohistochemistry[38], and one study reported the use of flow cytometry[34]. Notably, six of the fluorescence-based studies incorporated tissue-clearing before image acquisition to enhance visualization[23,24,33,35,37,38]. In addition, the study by Van De Looverbosch et al[33] employed computational mapping as an innovative strategy to quantify tumor invasion and spatial dynamics.

In terms of outcomes, invasion/migration patterns were often aligned with each study’s specific objectives. Some studies reported significant cellular invasion/migration, with GSCs dispersing over 100 μm from the NO borders, reaching up to 300 μm in some cases[33,38]. This level of penetration was associated with tumor aggressiveness and the presence of microtube formation[24,37]. When patient-derived GSCs were used, variations in invasion/migration depth and pattern were observed among cell lines[38].

The co-culture duration significantly influenced the extent of invasion. Time-course analyses demonstrated that tumor cell infiltration increased progressively over time, with more extensive invasion observed in models with prolonged culture periods[25,35]. In addition to duration, the maturity of the NOs also impacted the results - older organoids enabled deeper GSC integration, possibly due to the greater structural and microenvironment complexity they developed over time[23,35]. Moreover, one study evaluated the influence of cell seeding density, reporting that low-density GSCs resulted in sparse cellular dispersion, whereas high-density GSCs facilitated the formation of interconnected invasive streams, reflecting more aggressive infiltration patterns[14].

Some studies tested therapeutic interventions aimed at limiting GBM invasion into NOs. Treatment with Zika virus led to a significant reduction in tumor cell populations, even in models of other central nervous system tumors[34]. Another study evaluated 5-aminolevulinic acid-mediated phototherapy, which effectively eliminated infiltrative tumor cells while preserving surrounding normal neural tissue[25]. Further investigations examined the molecular mechanisms associated with invasiveness. The combined modulation of microRNAs miR-17-3p, miR-222, and miR-340 was found to inhibit tumor growth and promote cell death, thereby reducing invasion[36]. The induction of ciliogenesis also emerged as a promising strategy, as it promoted GSC differentiation and limited their capacity to infiltrate NOs[37]. Finally, one study highlighted the importance of standardizing and experimental approaches and introduced computational tools, including neural networks, to map cellular subpopulations and quantitatively assess tumor invasion patterns within the organoid models[33].

DISCUSSION

This systematic review synthesizes and compares methodologies and findings from studies using human-derived NOs to model GBM invasion. The integration of GBM cells into NOs enables the investigation of invasion dynamics within a neural microenvironment that more closely resembles in vivo conditions[39]. Furthermore, this review highlights various methodological approaches and strategies for analyzing tumor invasiveness. This approach provides a physiologically relevant platform for studying tumor progression and therapeutic responses, offering insights that may help bridge the gap between in vitro models and clinical observations.

Of the 377 initially identified articles, only 10 original research studies met all inclusion criteria, reflecting the relative novelty and technical complexity of this emerging research field. The selection and eligibility process required rigorous evaluation due to the significant heterogeneity across aim of the studies. A major challenge was the lack of standardized protocols for assessing invasion/migration; in many cases, invasion was not the primary outcome but a secondary or supporting analysis. These findings emphasize the need for standardized, well-documented protocols to enhance the reproducibility and reliability of findings in NO-based GBM models. The review focused on studies published from 2019 onward, aiming to capture recent and methodologically refined applications of NOs, whose use has notably expanded since their introduction in 2014[17].

Overall, the studies included in this review exhibited substantial methodological variability in the development of GBM models, particularly in terms of model type, cell source, and culture conditions. Suspension-based models were the most commonly used (70%), while spheroid-based models were employed in 30% of the studies. The preference for suspension systems may be attributed to their technical simplicity and greater reproducibility. However, spheroid models are often considered more physiologically relevant, as they better recapitulate the 3D architecture and cellular interactions of in vivo tumors[40]. Notably, one study directly compared both approaches and reported no statistically significant difference in organoid co-culture engraftment or tumor growth, regardless of whether GBM cells were introduced as suspension cells or as tumorspheres[25].

The number of GBM cells used to construct the models varied considerably across studies, a discrepancy likely influenced by the underlying tumor modeling strategies. Suspension-based systems often incorporate larger numbers of tumor cells and represent more dispersed populations, thereby maximizing direct contact with the NO and potentially enhancing early integration[40]. In contrast, spheroid-based models provide a more compact 3D structure that more closely mimics the spatial organization of in vivo tumors, although this structure may initially limit cell dissemination[41].

Beyond the number of cells, the invasive potential of GBM cells may also be related to the cell type selected. In the studies reviewed, 80% used patient-derived cells, with a notable predominance of GSCs. GSCs have been shown to possess higher invasive capacity compared to more differentiated tumor cells[42,43]. This methodological preference reflects a growing trend in the field toward the adoption of models that more accurately recapitulate the biological complexity of human tumors[42,44]. GSCs are widely recognized as a critical subpopulation within GBM, characterized by their capacity for self-renewal, multipotent differentiation, and pronounced phenotypic plasticity[45]. Additionally, accumulating evidence suggests their central role in therapeutic resistance, driven by both intrinsic mechanisms - such as elevated expression of drug efflux transporters and activation of DNA repair pathways – and extrinsic factors, including their privileged localization within hypoxic, protected niches inside the tumor[45,46]. Therefore, the prevalent use of GSCs in these models underscores their relevance in advancing our understanding of GBM invasion and treatment resistance.

In this context, the use of fluorescent labeling for real-time tracking of tumor cells is crucial, as it provides a powerful tool for quantifying and analyzing their invasive behavior within NOs models[47,48]. GFP is widely used due to its reliable expression and non-toxicity in live-cell imaging[49]. Enhanced variants, such as enhanced GFP, as reported by Van De Looverbosch et al[33], offer superior brightness and photostability for long-term or low-signal imaging conditions, compared to the wild-type protein. Nuclear-targeted forms such as histone-fused construct GFP, as used by Azzarelli et al[14], allow precise tracking of nuclear dynamics, mitotic events, and cell positioning within complex 3D structures[48]. The dominance of GFP and its enhanced variants reflects their robustness in live-cell tracking[49]. Additionally, red fluorophores such as mCherry and tdTomato further facilitate multiplex imaging and more refined spatial discrimination in co-culture assays, contributing to both enhanced visualization and quantitative monitoring of invasion and proliferation. These advances significantly strengthen the analytical power of the reviewed models[47-49].

Among the three studies utilizing GBM spheroids[23,25,35], only two specified the duration of tumor cell culture before co-culture with NOs (1-2 days)[23,35]. This critical parameter can affect spheroid compaction and internal organization, directly impacting the model’s ability to recapitulate in vivo tumor features. Longer pre-culture periods favor the development of physiologically relevant features including oxygen and nutrient gradients, hypoxic zones, and cellular heterogeneity, all essential for accurate tumor microenvironment modeling[50] and invasion studies[50,51]. Notably, none of these studies incorporated ECM components during spheroid formation, despite their crucial role in modulating cell adhesion, migration and signaling[50,52]. Integration of ECM elements could significantly improve the biological fidelity and structural complexity of these 3D tumor models[53].

The biological relevance of GBM invasion models also depends on the NOs used as brain-like platforms, and their structural and cellular complexity directly influences tumor cell behavior and patterns of tissue infiltration[39]. All of the selected studies employed hiPSCs to generate NOs, consistent with current scientific consensus, as highlighted by Pașca et al[18], who emphasized that the definition of organoids inherently involves their derivation from pluripotent stem cells, particularly hiPSCs, ensuring developmental plasticity necessary for self-organization into brain-like structures and physiological relevance foundation for modeling human neurodevelopment and disease[18] - notable exceptions exist in methodology. Van De Looverbosch et al[33] generated structures from hiPSC-derived NPCs that, despite being termed NOs, more accurately represent neural spheroids due to their restricted developmental potential. Although methodologically valid, this approach may reduce cellular heterogeneity and microenvironmental complexity compared to conventional NO models in which multipotent differentiation occurs dynamically during organoid formation[54]. It is also noteworthy that Bassot et al[36] reported the co-culture of hESCs and hiPSCs for NO generation, a strategy that produced organoids with characteristic cellular compositions and maturation patterns. Interestingly, Linkous et al[24] compared NOs derived independently from each pluripotent cell type and found that the source - hiPSCs vs hESCs - had no significant impact on tumor invasion outcomes. Both organoid types supported similar GBM growth and infiltration. Thus, while hiPSCs and hESCs may differ in their activity across key signaling pathways and intrinsic biological properties[55,56], these differences do not appear to substantially influence tumor-host interactions within the organoid model or affect the mechanisms of tumor invasion.

Although hiPSCs were the primary cell type for NO generation, considerable variability exists across protocols in terms of the initial cell number, culture duration, and final size of organoids, yet few studies report these variables in combination. For instance, organoids formed with 9000 or 10000 cells typically reached diameters between 500 and 1077 μm within 30 days[33,38], while those generated from 35000 cells showed smaller dimensions (500-700 μm), even after 60 days in culture[23]. This suggests that the initial number of cells does not directly correlate with the final organoid size. Only three studies reported the final dimensions of the organoids, highlighting a lack of standardization of these parameters. In contrast, culture duration, is widely recognized as a critical determinant of organoid maturation[57-59], with most studies using organoids at approximately 40 days of development[58,59]. Only one study systematically evaluated multiple time points (20, 40, and 60 days)[23]. The prevailing focus on early-stage organoids may limit the insights into later neurodevelopmental processes such as layer organization and synaptic network formation[60].

The NO age appears to influence the selection of characterization markers. During early stages, there was a predominance[14,24,25,34,35,38] of pluripotency and neural progenitor markers (nestin, musashi-1, sex determining region box 2, paired box protein 6 and TBR2) to indicate the onset of neurogenesis. All studies that performed cellular characterization[14,23-25,34-36,38], included neuronal markers to confirm NO differentiation at various stages. Beta-tubulin III, microtubule-associated protein 2 and BEARSKIN1 were commonly used to identify neuron layers, and TBR1 was associated with deeper cortical layers, reflecting both neuroepithelial progression and cellular diversity throughout organoid maturation[61,62]. Glial markers (oligodendrocyte transcription factor 2 and O4) appeared predominantly in more mature NOs, indicating increased cytoarchitectural complexity that more closely resembles in vivo brain tissue[63,64]. Additionally, the synaptic marker synapsin1 was detected in 60-day organoids[23], providing evidence of synapse formation in more advanced neural structures[65]. Although astrocytes are known to play a key role in promoting GBM cell model invasion[66], we found that only two studies[25,36], characterized astrocyte populations in NOs using GFAP and S100β markers. Additionally, only three studies[23,24,35] cultured NOs beyond 60 days, the approximate timeframe when astrocytes typically begin to emerge in these models[19].

In this systematic review, we found that 70% of the studies embedded NOs in ECM droplets during their generation. This approach may promote rapid tissue polarization and reorganization of the neuroepithelial architecture, supporting the emergence of radial glial cells and rosette structures, ultimately enhancing the structural similarity to the human brain[67]. Additionally, this strategy may also favor GBM cell invasion, as ECMs are known to provide biochemical and structural cues that continuously interact with GBM cells, facilitating and guiding their infiltration into brain tissue[68].

We also found that four studies[23,25,36,37] generated NOs using guided protocols based on dual-SMAD inhibition. This strategy is valuable, as it promotes the development of NOs that resemble the human cerebral cortex, which is the region most affected by GBM cells in the human brain[69,70]. In contrast, unguided NO protocols have the intrinsic potential to generate non-ectodermal brain cell types, such as microglia and endothelial cells[19,22], which have been shown to support and enhance GBM cell invasion[71]. However, none of the reviewed studies characterized these cell populations or explored their potential interactions with GBM cells.

GBM cells typically invade the brain by migrating along parenchymal and white-matter tracts, often exploiting the perivascular space, which provides essential oxygen and nutrients through its close association with blood vessels[72]. Moreover, immune cells such as microglia, neutrophils, and myeloid cells actively contribute to creating a tumor-permissive microenvironment that supports GBM invasion through various mechanisms[71]. In contrast, most current NO models lack key structural and cellular components of the adult brain, such as functionally mature vasculature and a fully developed immune compartment[73].

Moreover, GBM cells are thought to originate primarily from neural stem or progenitor cells rather than from fully differentiated glial cells. In the adult brain, these stem cells reside in deep neurogenic niches such as the subventricular zone and subgranular zone, from which they are believed to emerge and migrate outward to invade the cerebral cortex[74]. However, in GLICO models, tumor cells are applied to the surface of the organoids, limiting the ability of these systems to mimic the endogenous “inside-out” migration pattern observed in vivo. Collectively, these limitations hinder the capacity of current NO-based models to fully replicate the complex invasive behavior of GBM cells in the human brain, underscoring a critical gap in glioblastoma research.

Regarding the analysis of GBM invasion into NOs, approximately 90% of the studies analyzed in this review used the GLICO model[14,23-25,34-38], with a notable exception being the study by Van De Looverbosch et al[33], which did not employ a true NO in its experimental design. A key observation in this review was the methodological consistency across studies in both co-culture conditions and invasion analysis. The widespread use of low-adhesion surface plates and fluorescence-based imaging techniques facilitated interaction between tumor and NO cells and enabled accurate tracking of tumor cells within the 3D microenvironment, respectively. However, despite this consistency, there was no clear consensus regarding the duration of co-culture. Most studies reported co-culture periods ranging from 10 and 20 days[23,24,33,34,37], while two studies extended the duration between 40 and 90 days[25,35]. These variations likely reflect differences in experimental goals, the maturation stage of the organoids, or the complexity of the tumor model used. Shorter co-culture periods tend to be employed to study early events such as cell adhesion, migration, and the initiation of invasion, making them useful for rapid phenotypic screening of invasive phenotypes or preliminary therapeutic testing.

Given that the analysis of tumor invasion is the central focus of this review, five studies explored strategies to optimize the GLICO model by manipulating variables such as cell density, organoid maturation stage, the inclusion of ECM components, and the type of tumor model employed[14,23,25,33,35]. These efforts aimed to improve experimental conditions to more accurately replicate the tumor microenvironment. Four additional studies incorporated therapeutic interventions within the context of tumor invasion[24,34,36,38]. Ferreira et al[34] assessed the oncolytic effects of the Zika virus, reporting a significant reduction in tumor cell populations post-infection, particularly within infiltration zones. Bassot et al[36] observed that modulation of specific microRNAs led to a decrease in GBM invasiveness. Similarly, Goranci-Buzhala et al[37] found that inhibition of NEK2-KD, a kinase involved in cell division, promoted tumor cell differentiation and limited invasion. Linkous et al[24] reported that conventional chemo- and radiotherapeutic agents (temozolomide, carmustine, and ionizing radiation) effectively reduced tumor infiltration. Additionally, Krieger et al[38] compared the invasive potential of four patient-derived GSC lines, revealing intertumoral differences in terms of penetration depth. Collectively, these findings reinforce the potential of the GLICO model as a versatile platform for both personalized therapeutic testing and mechanistic studies aimed at limiting tumor invasion.

This review presents some limitations that warrant consideration. The relatively small number of included studies restricts the breadth and generalizability of the findings. Moreover, substantial methodological heterogeneity was observed across studies, including variations in cell sources, differentiation protocols, organoid maturation stages, tumor model types, and techniques used to assess tumor invasion/migration. This variability hindered direct comparisons between studies. These limitations highlight the need for standardized protocols to improve reproducibility and facilitate future translational applicability.

To address these issues, targeted strategies should be considered. For example, the establishment of consensus guidelines, similar to the Minimum Information About a Microarray Experiment initiative in genomics, could promote transparency and uniformity in reporting experimental parameters such as cell density, culture duration, and invasion assessment methods. Collaborative efforts through multi-institutional working groups or consortia would facilitate the development of such guidelines, ensuring that model validation, organoid characterization, and invasion quantification follow agreed-upon standards. Furthermore, benchmarking organoid-based GBM models against other established in vitro or ex vivo systems, such as slice cultures or 3D bioprinted brain matrices, may help define their relative strengths and limitations. The use of inter-laboratory reproducibility studies, including blinded sample analyses, can also play a key role in refining methodologies and identifying critical parameters affecting variability. Finally, adapting standardization practices from other fields, such as the use of reference materials, centralized protocols, or open-access databases for protocol sharing, could significantly enhance consistency and comparability across studies.

In parallel, future studies should aim to deepen the analysis and interpretation of invasion-related findings, going beyond descriptive comparisons. For instance, differences in depth of invasion across studies may reflect not only intrinsic tumor cell properties, such as stemness, proliferation rate, or epithelial-mesenchymal transition-like phenotypes, but also technical factors including NO maturity, ECM composition, or co-culture duration. Similarly, the choice of invasion quantification method (e.g., 2D fluorescence area, 3D volumetric reconstructions, radial intensity profiling) strongly influences the sensitivity and spatial resolution of the analysis. Yet, few studies directly compare the strengths and limitations of these approaches. A more critical evaluation of the methodological tools, including their reproducibility, scalability, and ability to capture invasion dynamics over time, would enhance the interpretative value of the results. Finally, the convergence or divergence of findings across models should be interpreted in light of both biological heterogeneity and experimental variability in the understanding of GBM invasion.

CONCLUSION

This systematic review analyzed studies that employed human stem cell-derived NOs as platforms to investigate GBM migration and invasion. Although the number of relevant publications over the past five years remains limited, GLICO co-culture models have shown considerable potential for exploring the cellular and molecular mechanisms underlying tumor progression. The evidence suggests that the structural and functional complexity of organoids, when combined with GSCs, provides a humanized brain microenvironment that closely mimics in vivo conditions, enabling more representative analyses of tumor aggressiveness. Therefore, these models have shown promise for personalized therapeutic testing and drug screening. Advancements such as the integration of functional vasculature, immune system components, and patient-specific genetic backgrounds are expected to further enhance the utility of organoid-based systems for studying tumor-host interactions, therapeutic resistance, and the identification of novel treatment targets. However, methodological variability across studies presents a key limitation, hindering reproducibility and meaningful cross-study comparisons. Therefore, future efforts should prioritize the standardization of organoid generation protocols and invasion assessment, strengthening the translational applicability of these models.