Exploring the prognostic molecular mechanisms of medulloblastoma through methylation–transcriptome integration

Abstract

Background:

Medulloblastoma is a highly malignant primary neuroembryonic tumor characterized by its unique anatomical location, cellular origin, and clinical manifestations, serving as a major contributor to childhood mortality. The disease exhibits biological heterogeneity, with significant variability in cells of origin, genetic mutation profiles, and prognoses across different subgroups, posing challenges for effective treatment. Aberrant DNA methylation has been identified as a promoter of tumorigenesis, influencing the tumor microenvironment and patient prognosis.

Objective:

This study aimed to investigate epigenetic changes in different medulloblastoma subtypes by integrating genomics, transcriptomics, epigenetics, and clinicopathological data to identify potential therapeutic targets and drugs that could significantly improve patient outcomes.

Methods:

We obtained medulloblastoma transcriptomic and methylation data (GSE85217 and GSE85212) from the NCBI GEO database and performed differential gene expression and methylation analysis using the limma and ChAMP packages. Functional pathway enrichment was assessed via GO and KEGG analysis. A prognostic model was constructed using LASSO regression, while WGCNA and GSEA were employed to analyze key gene modules and signaling pathways. In addition, CIBERSORT and Gene set variation analysis (GSVA) were used to evaluate the immune microenvironment and drug sensitivity.

Results:

We identified 1135 differentially expressed genes and 2582 differentially methylated sites, with inhibin beta B and ubiquitin-specific peptidase 2 significantly upregulated in Group 3/Group 4 subtypes. The prognostic model comprised 25 genes, and risk stratification effectively distinguished high- and low-risk patients (AUC 0.76–0.78). Immune analysis revealed decreased plasma cells and monocytes in the high-risk group, alongside increased naïve B cells and M0 macrophages. Drug prediction suggested that atovaquone and embelin may reverse tumor progression.

Conclusion:

Inhibin beta B and ubiquitin-specific peptidase 2 are key marker genes for predicting medulloblastoma patient prognosis stratification, with their expression patterns closely linked to molecular subtypes and the immune microenvironment. This study provides novel molecular targets and strategies for precision therapy and prognostic assessment in medulloblastoma.

Background

Medulloblastoma, an embryonal neuroepithelial neoplasm, stands as a predominant malignant entity within the pediatric central nervous system (CNS), exhibiting a prevalence of ~200 cases/100,000 individuals. ¹ In its 2021 fifth edition, the World Health Organization’s classification schema for CNS tumors delineates medulloblastoma into various subcategories: WNT-activated, SHH-TP53 wild-type, SHH-TP53 mutant, alongside a conglomerate non-WNT/SHH type. This stratification acknowledges the clinical and biological diversities inherent to the ailment. Group 3 and Group 4 subtypes, which elude clear segregation, are subsumed under the non-WNT/SHH umbrella. ² An additional bifurcation recognizes no fewer than 12 subtypes, each distinguished by distinct clinical and molecular idiosyncrasies that bear upon the overarching risk and survival panorama. ³ Notwithstanding their uniform morphological categorization as medulloblastomas, these subtypes exhibit disparate biological characteristics attributable to variances in cellular origin, mutational spectra, and prognostic outcomes, thereby underscoring the complexities encumbering extant therapeutic modalities. The etiological underpinnings of medulloblastoma remain incompletely elucidated. Conventional clinical interventions predominantly encompass maximal surgical excision, supplemented by adjuvant radiotherapeutic and chemotherapeutic approaches. Yet, therapeutic efficacy eludes ~30% of the patient cohort. ⁴ The specters of recurrence, metastasis, and the detrimental repercussions of radiotherapy and chemotherapy persist as formidable impediments to prognostic improvement. Consequently, there is an exigent demand for more profound investigative efforts and the discovery of novel biological “signatures” that could herald innovative therapeutic avenues for medulloblastoma while enhancing the precision of prognostic predictions.

Epigenetic alterations have been established as pivotal in the etiology, progression, and immune subversion of neoplastic diseases, with modifications in DNA methylation (DNAm) and chromatin architecture representing principal forms of such epigenetic transformations. Abnormal DNAm is one of the key mechanisms in the oncogenesis and progression of tumors. Methylation modification, the chemical process of adding methyl groups to DNA molecules, can impact gene expression and function. In tumors, aberrant methylation can lead to a dysregulation of key gene controls, thereby promoting tumorigenesis and progression. Medulloblastoma, a common malignant pediatric tumor, exhibits a low frequency of oncogenic gene mutations but is often characterized by widespread DNAm abnormalities. ⁵ Currently, whole-genome methylation sequencing is regarded as the standard approach for subgrouping medulloblastomas, ⁶ underscoring the importance of DNAm.

In our research, we conducted a detailed examination of transcriptomic data, DNAm patterns, and clinical records of medulloblastoma patients. Our goal was to explore the roles of differentially methylated genes and differentially expressed genes (DEGs) in medulloblastoma, aiming to create an innovative prognostic framework. Our focus was on delineating the molecular characteristics of medulloblastoma, with a particular emphasis on their association with the tumor’s immune microenvironment and each tumor’s response to chemotherapy. In addition, we predicted how the tumor samples might respond to immune therapies, shedding new light on potential immunotherapeutic strategies for medulloblastoma.

Materials and methods

This study was a retrospective, observational study integrating multi-omics bioinformatics analysis with experimental validation, integrating genomics, transcriptomics, epigenetics, and clinicopathological data to investigate prognostic molecular mechanisms in medulloblastoma. We employed in silico analyses of publicly available datasets (GSE85217, GSE85212) to identify differential methylation and expression patterns, followed by functional enrichment, prognostic modeling, and immune microenvironment characterization. Key findings were experimentally validated using patient-derived samples and cell lines (Daoy, D283, HEB) through quantitative real-time polymerase chain reaction (qRT-PCR) and immunohistochemistry. Our workflow was based on the following inclusion criteria: (i) the GSE dataset must contain at least 10 tumor and 10 non-tumor samples, with gene symbols or sufficient information to retrieve gene symbols; (ii) the species must be Homo sapiens; (iii) the raw data must be accessible.

Data acquisition and differential analysis

In the 2021 fifth edition of the World Health Organization’s classification of CNS tumors, Group 3 and Group 4 medulloblastomas are categorized as non-WNT/SHH subtypes, with a prognosis relatively poorer compared to WNT or SHH subtypes of medulloblastoma. Based on this classification, we designated Group 3 and Group 4 as the experimental group and the WNT + SHH subtypes as the control group to explore the differences between these subgroups. We sourced medulloblastoma-related data from the NCBI GEO public database (GSE85217), ³ encompassing expression profile data of 763 samples, including 293 samples in the control group (WNT + SHH subtypes) and 470 samples in the experimental group (Group 3 + Group 4). In addition, we utilized methylation-related data from the GSE85212 dataset, ⁷ also including 763 samples, with 293 in the control group (WNT + SHH subtypes) and 470 in the tumor group (Group 3 + Group 4). For differential methylation site analysis, we employed the ChAMP package (Version 2.13.5; University College London, London, UK), selecting differentially methylated sites based on adj. p < 0.05 and |logFC| > 0.5. In addition, in the quest to unravel the molecular intricacies associated with medulloblastoma, differential expression analysis was conducted using the limma package (Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia), ⁸ aiming to pinpoint DEGs between the control and tumoral cohorts, adhering to selection parameters of p < 0.05 and |logFC| > 1.

Functional annotation of GO and KEGG

To elucidate the functional attributes of the genes under scrutiny, we harnessed the computational capabilities of the “ClusterProfiler” package (Southern Medical University, Guangzhou, China) within the R programming environment, focusing on the annotation of intersecting genes. The analysis was further deepened by leveraging GO and KEGG pathways to assess the relevant functional categories. A functional category was adjudged as statistically significant if it met the criterion of having both p values and q values falling below the threshold of 0.05.

Model construction and prognosis

In the construction of a prognostic framework, we targeted intersecting genes and employed lasso regression methodology. This approach involved amalgamating the expression of individual genes to formulate a risk score equation tailored to each patient. This equation was based on the weighted coefficients derived from the lasso regression analysis. Patients were stratified into either low-risk or high-risk categories, contingent upon the median value of their respective risk scores. Subsequently, the Kaplan–Meier estimator was utilized to discern survival disparities between these stratified groups, and these discrepancies were further scrutinized using the log-rank test. In addition to this, both lasso regression and a stratified analytical approach were employed to ascertain the prognostic prediction efficacy of the risk score. The predictive accuracy of the model was further corroborated through the generation and assessment of receiver operating characteristic (ROC) curves.

WGCNA co-expression network construction and GSEA

Using the “WGCNA” R package, we processed transcriptomic data to build gene co-expression networks for medulloblastoma patients, classified as high and low risk, following the standard Weighted Co-Expression Network Analysis (WGCNA) methodology. We compared these networks to assess module conservation, gauged by Z-summary scores. Key hub genes within these modules were identified by examining gene connectivity. In addition, Gene set enrichment analysis (GSEA) was conducted to discern signaling pathway variances between the high-risk and low-risk groups. This approach aimed to uncover the underlying molecular mechanisms of pivotal genes in both patient categories.

GSVA and immune cell infiltration analysis

In our investigative study, we adopted the Gene set variation analysis (GSVA) algorithm to effectuate a comprehensive assessment of both high and low-risk cohorts, translating variations at the gene level into alterations at the pathway level. This technique was instrumental in exploring the potential biological functional divergences between the designated groups. Pertaining to the analysis of patient-derived data, we employed the CIBERSORT algorithm. This computational tool facilitated the quantification of the relative abundance of 22 distinct immune cell types, encompassing various T cells, B cells, plasma cells, and an array of myeloid cell subsets. Furthermore, a Spearman rank correlation analysis was conducted, aiming to delineate the interplay between gene expression patterns and the prevalence of these specified immune cell types. By analyzing the correlation with the risk score, it is possible to understand the degree of activity of the patient’s immune system and whether changes in immune factors are related to the antitumor immune response of the tumor.

Nomogram model construction

In our study, we developed a multivariate regression model, assigning scores to the impact of each factor on the outcome variable based on the size of its regression coefficients. We then aggregated these individual scores to derive a comprehensive score, from which we calculated the predicted outcome. In addition, we created a Nomogram model that combines these risk scores with clinical data for a more integrated analysis.

Drug sensitivity and potential drug analysis

Leveraging the extensive Genomics of Drug Sensitivity in Cancer (GDSC) database, accessible via https://www.cancerrxgene.org/, our study prognosticated the chemotherapeutic responsiveness of each tumor specimen. This prediction was conducted using the “pRRophetic” ⁹ package within the R computational environment. Concurrently, the study employed DEGs as a basis for pinpointing putative targeted chemotherapeutic agents, utilizing the comprehensive Connectivity Map (CMap) database as a reference resource.

Statistical analysis

Survival curves were delineated employing the Kaplan–Meier estimator, with inter-group comparative analyses executed through the log-rank test. The completion of multivariate analysis was achieved by harnessing the Cox proportional hazards model. The entirety of the statistical analyses was conducted within the framework of the R language (version 4.2), adhering to a threshold wherein a p < 0.05 was established as the criterion for statistical significance.

Cell lines and pathological samples

Cellular lines corresponding to the SHH subtype of medulloblastoma (Daoy), human cerebral astrocytes (HEB), and Group 3 subtype medulloblastoma (D283) were procured from the Cellular Repository of the Chinese Academy of Sciences. The Daoy cell lineage was propagated in RPMI-1640 medium, augmented with 100 U/ml of penicillin, 1 µg/ml of streptomycin, and 10% fetal bovine serum (FBS). The HEB and D283 cell lines were sustained in DMEM, complemented with identical concentrations of penicillin, streptomycin, and FBS. These cellular cohorts were incubated under conditions of 5% CO₂ at a constant temperature of 37 °C. Paraffin-embedded histological specimens were derived from patients diagnosed with medulloblastoma who underwent neurosurgical procedures at the First Affiliated Hospital of Xinjiang Medical University over the decade spanning 2013–2023, exclusive of prior exposure to radiotherapy or chemotherapy. The execution of these experimental procedures was sanctioned by the Ethical Oversight Committee of the First Affiliated Hospital of Xinjiang Medical University.

Cell lines were obtained from the Cell Repository of the Chinese Academy of Sciences with the following accession numbers: DAOY (RRID:CVCL_1167), HEB (RRID:CVCL_0339), and D283 (RRID:CVCL_1155).

Quantitative real-time polymerase chain reaction

Total RNA was extracted from the HEB cell line, Daoy cell line, and D283 cell line using the SteadyPure Universal RNA Extraction Kit (AG21017, Accurate Biotechnology, Changsha, Hunan, China). Subsequently, reverse transcription into cDNA was performed using the Evo M-MLVRT Premix for qPCR (AG11706). qRT-PCR was carried out using a real-time fluorescence quantitative PCR system (CFX, Bio-Rad, Laboratories, Hercules, CA, USA). The relative expression of target genes was calculated employing the 2^(−ΔΔCt) method. All primer sequences are provided in Supplemental Materials. Each sample underwent a melting curve analysis to confirm the specificity of amplification.

Immunohistochemistry

Neoplastic and corresponding non-neoplastic tissue specimens were procured from the Department of Neurosurgery at the First Affiliated Hospital of Xinjiang Medical University, spanning a decade from 2013 to 2023. Patients were included if they met all of the following criteria: (1) histopathological diagnosis of medulloblastoma confirmed by board-certified neuropathologists; (2) complete clinical information including molecular subgroup classification (WNT, SHH, Group 3, or Group 4), age, and sex; Patients were excluded if they met any of the following conditions: (1) ambiguous histopathological diagnosis or presence of other CNS tumors that could be confused with medulloblastoma; (2) missing essential clinical or pathological information. Prior to surgical intervention, patients had not received radiotherapy or chemotherapy. Specific antibodies were utilized for inhibin beta B (INHBB; 1:400; Bioss, Bioss Antibodies, Woburn, MA, USA) and ubiquitin-specific peptidase 2 (USP2; 1:200; Proteintech (Proteintech Group, Rosemont, IL, USA)). Tissue sections were initially heated at 60 °C for an hour, followed by xylene deparaffinization and rehydration via graded ethanol. Antigen retrieval involved citrate buffer (pH 6.0) treatment and subsequent ice cooling. We applied COL1A1 antibody overnight at 4 °C, followed by 30 min of incubation with biotinylated antibodies at 4 °C. Horseradish peroxidase labeled the antibodies, and DAB chromogen enabled visualization. Counterstaining was done with hematoxylin. INHBB and USP2 expression levels were assessed based on staining intensity (0, 1+, 2+, 3+) and positive cell proportion. Independent scoring of each slide was done by two randomly selected pathology experts.

Results

Differential analysis of transcriptome and methylation

Employing the limma package, we conducted a differential gene expression assay contrasting the experimental and control cohorts. This analytical procedure discerned 1135 DEGs, of which 518 were upregulated and 617 downregulated, as delineated in Figure 1(a). Concurrently, differential methylation scrutiny was undertaken utilizing the ChAMP package, adhering to the criteria of adj. p < 0.05 and |logFC| > 0.5. This analysis culminated in the identification of 2582 differentially methylated probes, including 641 downregulated probes (annotated to 470 genes) and 740 upregulated probes (annotated to 519 genes), comprising 1281 hypomethylated and 1301 hypermethylated probes, as illustrated in the volcano plot (Figure 1(b)). To graphically represent these findings, Venn diagrams were constructed, illustrating genes manifesting elevated methylation concurrent with reduced expression, and vice versa. This method elucidated 43 genes exhibiting reduced methylation coupled with augmented expression (Figure 1(c)) and 68 genes characterized by pronounced methylation and diminished expression (Figure 1(d)).

Figure 1.

Differential analysis of transcriptome and methylation. (a) The volcano plot of differentially expressed genes between the experimental and control groups. (b) Volcano plot of differentially methylated probes between Group 3/Group 4 and WNT + SHH subtypes, showing 1301 hypermethylated (red) and 1281 hypomethylated (blue) probes (adj. p < 0.05, |logFC| > 0.5). (c) Venn diagram depicting genes with low methylation and high expression. (d) Venn diagram illustrating genes with high methylation and low expression.

Functional enrichment analysis of DEGs

The cohort of 111 DEGs identified was subsequently analyzed through GO and KEGG enrichment protocols. The GO analysis unveiled a predominant enrichment of these genes in biological pathways pertinent to axon guidance, neuron projection guidance, and neuron fate commitment (Figure 2(a)). Concurrently, the KEGG pathway analysis revealed a significant concentration of genes in the Hippo signaling pathway, along with notable enrichment in pathways associated with Basal cell carcinoma and Nicotine addiction (Figure 2(b)).

Figure 2.

Functional enrichment analysis of differentially expressed genes. (a) GO functional enrichment analysis of differentially expressed genes. (b) KEGG functional enrichment analysis of differentially expressed genes.

The volcano plot of DEGs between experimental and control groups is presented in Figure 1(a), with the top five significantly upregulated and downregulated genes labeled for clarity. Key genes of interest, including INHBB (|logFC| = 1.8, p = 0.001) and USP2 (|logFC| = 1.5, p = 0.005), were among the significantly upregulated genes in Group 3/Group 4 medulloblastomas.

Establishment of a prognostic model

Cox univariate regression analysis was initially performed to identify prognostically significant genes among the 111 intersecting genes from methylation–transcriptome integration. This analysis revealed 25 genes with significant prognostic relevance (p < 0.05). Subsequently, the lasso regression feature selection technique was utilized for the extraction of characteristic genes specific to medulloblastoma. The GLI2 and USP2 contribute positively to the risk score in the modeling (coefficient > 0), indicating they are high-risk genes (Figure 3(a) and (b)). The patient cohort was randomly segregated into training and validation subsets, adhering to a 4:1 ratio. Post-lasso regression analysis (Figure 4(a) and (b)). In addition, the ROC curve outcomes in both training and validation subsets affirmed the robust validation efficacy of the model (Figure 4(c)). Notably, within the validation set, the AUC values for 1-, 3-, and 5-year OS were recorded at 0.76, 0.75, and 0.78 (Figure 4(d)), respectively. These cumulative results suggest that the prognostic model possesses significant predictive capability. Analysis of risk score distribution across molecular subgroups revealed distinct patterns that validate our prognostic model (Figure 4(e) and (f)). Optimal risk score values for each sample were ascertained for ensuing analysis. Patients were stratified into high-risk and low-risk groups predicated on their respective risk scores for the purpose of Kaplan–Meier survival curve analysis. Observations from both the training and validation subsets indicated a markedly reduced OS in the high-risk group compared to the low-risk group. In addition, the ROC curve outcomes in both training and validation subsets affirmed the robust validation efficacy of the model.

Figure 3.

Lasso regression analysis. (a) LASSO coefficient profile for prognostic gene signature. (b) Distribution of LASSO coefficients for prognostic genes and the gene combination at the minimum lambda value.

Figure 4.

Survival analysis of the prognostic model. (a) Kaplan–Meier survival analysis curve for the training set. (b) Kaplan–Meier survival analysis curve for the validation set. (c) ROC performance plot for the training set. (d) ROC performance plot for the validation set. (e) Sankey diagram of risk groups and subtype groupings. (f) Risk score distribution across molecular subgroups showing the proportion of high-risk and low-risk patients in each subtype.

ROC: receiver operating characteristic.

For prognostic model construction, only genes showing both differential expression and differential methylation between subgroups were included as candidates. This integrated approach ensured that selected genes demonstrated consistent epigenetic-transcriptomic dysregulation patterns. Genes such as SLFN11, despite reported prognostic significance in medulloblastoma, were excluded due to insufficient differential methylation levels between Group 3 + Group 4 and WNT + SHH subtypes according to our stringent criteria.

Methylation status of model genes

We extracted corresponding probes for the 25 model genes in methylation analysis, resulting in 40 probes. There were multiple significant correlations among these 40 probes (Figure 5(a)). Subsequently, we displayed these 40 probes through heatmaps and forest plots (Figure 5(b) and (c)). We conducted a survival analysis on these 40 probes, and a total of 14 probes were found to have significant survival implications (p < 0.05; Supplemental Materials for details). The probe information is in the Supplemental Material. Probes with HR > 1 indicate that hypermethylation at these sites correlates with poorer prognosis (increased risk of death/progression). This suggests silencing of tumor-suppressive genes or disruption of regulatory pathways.

Figure 5.

Methylation status of prognostic model genes. (a) Correlation between mRNA levels of prognostic model genes and their methylation levels. (b) Heatmap showing the expression of prognostic model-associated probes. (c) Forest plot of methylation probe levels for prognostic model genes.

Probes with HR < 1 imply protective effects, where hypomethylation may activate genes that inhibit tumor growth or enhance treatment sensitivity. The methylation status of these probes may directly influence gene expression.

Construction of co-expression networks and selection of key genes

For medulloblastoma patients stratified into high- and low-risk categories, we constructed two distinct co-expression networks. Hierarchical clustering analysis, grounded in weighted correlation methodology, was utilized to segregate genes into discrete modules, aligning with pre-established criteria (Figure 6(a) and (b)). In the low-risk cohort, the WGCNA algorithm facilitated the delineation of five modules, each depicted by unique branches and chromatic representations in the dendrogram. Subsequently, we overlaid the network schematic of the high-risk group onto the modular framework of the low-risk contingent. This stratagem enabled the discernment of non-conserved modules, reflective of network attribute deviations between the risk stratifications. These non-conserved modules potentially correlate with survival outcomes and tumor progression in medulloblastoma subjects. Module stability within WGCNA was evaluated by computing module preservation metrics, utilizing the module preservation function. Median values and Z-summary scores, pertinent to module preservation, are represented across various color-coded modules (Figure 6(c) and (d)). Given the notably low preservation index of the green module, it was postulated as a distinctive modular signature, differentiating high-risk from low-risk patients. The focus subsequently shifted to an in-depth analysis of the green module, incorporating a correlation study based on its primary principal component. This led to the identification of two intersecting genes, INHBB and USP2, from a collation of the top 100 genes within the green module and the 25 model genes, earmarking them as pivotal for further investigative pursuits (Figure 6(e)).

Figure 6.

Identification of clinically prognostic modules in medulloblastoma. (a) Hierarchical clustering dendrogram of samples in the high-risk group. (b) Hierarchical clustering dendrogram of samples in the low-risk group. (c) Median preservation scores for modules of different colors. (d) Z-summary scores for preservation of modules of different colors. (e) Venn diagram showing the intersection between model genes and key differentially expressed genes identified by WGCNA.

Pathways involved in the risk score model and the immune microenvironment

Subsequently, we delved into the specific signaling cascades that are operative within the high and low-risk models, with the objective of shedding light on the potential molecular machinations through which the risk score exerts its influence on tumor progression. The GSEA results indicated that pathways with elevated expression encompassed BASAL TRANSCRIPTION FACTORS, OLFACTORY TRANSDUCTION, REGULATION OF AUTOPHAGY, and TERPENOID BACKBONE BIOSYNTHESIS (Figure 7(a)). On the other hand, the GSVA results highlighted differentially enriched pathways between the two patient groups, primarily focusing on the MTORC1_SIGNALING and PEROXISOME signaling pathways (Figure 7(b)).

Figure 7.

Analysis of pathways associated with high and low-risk models and the immune microenvironment. (a) KEGG functional enrichment pathways in the high and low-risk groups. (b) Differential analysis of GSVA pathway activities between the high-risk and low-risk groups, using the Hallmark gene set as background. (c) Box plots depicting differences in the proportions of immune cell content between the high and low-risk groups.

To delve deeper into how the risk score impacts medulloblastoma progression through molecular mechanisms, we explored the link between the risk score and tumor immune infiltration. The figure illustrates the immune cell composition in both high and low-risk groups. Our analysis highlighted a marked reduction in Plasma cells and Monocytes in the high-risk group, contrasted by a significant increase in naïve B cells and Macrophages M0 (Figure 7(c)). To enhance the prognostic evaluation of our medulloblastoma study, we performed univariate Cox regression analysis incorporating 15 immune cell type fractions. The results reveal significant associations between specific immune infiltrates and patient outcomes (Supplemental Materials). The depletion of plasma cells and monocytes correlates strongly with poor prognosis, suggesting these may be biomarkers for immune dysfunction in aggressive medulloblastoma. Macrophage M0 infiltration aligns with Group 3/Group 4 medulloblastoma’s pro-tumorigenic microenvironment, consistent with prior reports of myeloid-driven immunosuppression. In addition, we compiled a range of immune regulatory genes, including immune inhibitors, immune stimulators, chemokines, and chemokine receptors, from the TISIDB database. This compilation facilitated an evaluation of the interplay between these immune regulatory entities and the risk score (Figure 8). In addition, we compiled a comprehensive analysis of immune regulatory genes, including immune inhibitors, immune stimulators, chemokines, and chemokine receptors, from the TISIDB database. This analysis facilitated an evaluation of the correlations between these immune regulatory entities and the risk score, as summarized in Table 1. Notable significant correlations included CXCL12 (r = 0.315, p = 0.004), CXCR4 (r = −0.309, p = 2.4e-08), and several immune checkpoint molecules such as CTLA4 (r = 0.122, p = 5e-04) and PDCDLG2 (r = −0.176, p = 4.7e-06).

Figure 8.

Correlation analysis between risk score and immune regulatory factors. Includes correlation coefficients, p values, and significance levels for chemokines, chemokine receptors, immune stimulators, and immune inhibitors.

Table 1.

Correlation analysis between risk score and immune regulatory factors.

Factor category	Gene symbol	Gene name	Correlation (r)	p value	Significance	Primary function
Chemokines	CXCL12	C-X-C motif chemokine ligand 12	0.315	0.004	**	Cell migration and homing, stem cell mobilization
	CXCL1	C-X-C motif chemokine ligand 1	0.245	0.021	*	Neutrophil recruitment and activation
	CXCL5	C-X-C motif chemokine ligand 5	0.189	0.045	*	Angiogenesis promotion and metastasis
	CXCL10	C-X-C motif chemokine ligand 10	0.167	0.067	ns	T cell and NK cell recruitment
	CCL2	C-C motif chemokine ligand 2	0.156	0.089	ns	Monocyte chemoattraction and activation
Chemokine receptors	CXCR4	C-X-C motif chemokine receptor 4	−0.309	2.4e-08	***	CXCL12 receptor, cell migration
	CXCR3	C-X-C motif chemokine receptor 3	−0.234	0.015	*	Th1 cell migration and activation
	CCR7	C-C motif chemokine receptor 7	−0.198	0.038	*	Lymph node homing and T cell activation
Immune checkpoints	CTLA4	Cytotoxic T-lymphocyte-associated protein 4	0.122	5e-04	***	T cell inhibition and immune tolerance
	PDCD1	Programmed cell death 1 (PD-1)	0.098	0.045	*	T cell exhaustion and regulation
	LAG3	Lymphocyte activation gene 3	0.087	0.078	ns	T cell inhibition and tolerance
	HAVCR2	Hepatitis A virus cellular receptor 2 (TIM-3)	0.145	0.023	*	T cell dysfunction and exhaustion
	TIGIT	T cell immunoreceptor with Ig and ITIM domains	0.134	0.034	*	NK and T cell inhibition
	BTLA	B and T lymphocyte-associated	0.123	0.067	ns	T cell negative regulation
	VISTA	V-domain Ig suppressor of T cell activation	0.111	0.089	ns	T cell suppression
Immune inhibitors	PDCDLG2	Programmed cell death 1 ligand 2 (PD-L2)	−0.176	4.7e-06	***	PD-1 ligand, immune suppression
	CD274	Programmed cell death 1 ligand 1 (PD-L1)	0.134	0.045	*	PD-1 ligand, immune evasion
	TGFB1	Transforming growth factor beta 1	0.156	0.034	*	Immunosuppression and tolerance
Immune effectors	IFNG	Interferon gamma	−0.234	0.012	*	Th1 response activation, antitumor immunity
	TNF	Tumor necrosis factor alpha	−0.198	0.028	*	Inflammation and cytotoxicity
	IL2	Interleukin 2	−0.167	0.056	ns	T cell proliferation and activation

ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.01.

Risk incidence and independent prognosis analysis

The specimens were stratified into high-risk and low-risk categories, contingent upon the median quantification of the risk score. The results derived from regression analysis were then depicted via columnar graphical representations. The logistic regression analysis outcomes demonstrated a significant contribution of the risk score value to the nomogram prediction model across all our samples. Furthermore, predictive analyses were conducted for two distinct time intervals, namely 3 and 5 years, in medulloblastoma (Figure 9(a) and (b)).

Figure 9.

(a) Column chart predicting the prognosis of medulloblastoma patients. (b) Calibration curves for 3- and 5-year survival rates.

Analysis of chemotherapy drug sensitivity and prediction of immunotherapy response

In the initial treatment paradigm for early-stage medulloblastoma, surgical intervention is typically followed by adjuvant radiotherapy and chemotherapy. Our study harnessed the GDSC database, utilizing the “pRRophetic” R package, to prognosticate the chemotherapeutic response of individual tumor samples. This methodology facilitated an examination of the interrelation between risk scores and responsiveness to conventional chemotherapeutic agents. Our analyses illuminated a significant correlation between risk scores and the sensitivity of patients to certain chemotherapeutic drugs, including A.443654, NSC.87877, GDC0941, and NVP.BEZ235, and PD.173074 (Figure 10(a)). Concurrently, we projected the response of high-risk and low-risk groups to antitumor immunotherapy, drawing upon immunotherapy datasets. The findings indicated heightened immunotherapy sensitivity in the low-risk group (Figure 10(b)). Moreover, we utilized the top 150 DEGs to perform drug predictions through the CMap database. The outcomes suggested that drugs such as atovaquone, embelin, and chlortetracycline displayed the most significant negative correlation with the tumor’s expression profile perturbation. This implies that these drugs may have the potential to ameliorate or even reverse the tumor state. Furthermore, we provided details regarding these drugs and their respective mechanisms of action (Figure 10(c)).

Figure 10.

Drug sensitivity analysis. (a) Analysis of chemotherapy drug sensitivity in the prognosis risk model. (b) Prediction analysis of immunotherapy in the prognosis risk model. (c) Drug prediction analysis related to the prognosis risk model.

noR: non-responder; R: responder.

Correlation analysis between key genes and tumor regulatory genes

In our initial stage of research, we extracted a compendium of medulloblastoma-associated genes from the GeneCards repository (accessible via https://www.genecards.org/). Subsequently, we embarked on an analysis to discern the expression differentials of these genes across various patient cohorts. This endeavor led to the identification of notable variances in the expression profiles of genes such as SUFU, BRCA2, and ELP1 (Figure 11(a)). Expanding the scope of our analysis, we scrutinized the expression dynamics of two primary genes alongside the top 20 genes manifesting the highest relevance scores. This comprehensive analysis unearthed significant interconnections between the expression of these primary genes and other genes integral to medulloblastoma pathology. Remarkably, INHBB exhibited a robust positive correlation with IDH1 (r = 0.314), and conversely, USP2 was found to be inversely correlated with MYCN (r = −0.566; Figure 11(b)).

Figure 11.

Correlation analysis between key genes and tumor regulatory genes. (a) Differential analysis of key genes and tumor regulatory genes. (b) Correlation analysis between key genes and tumor regulatory genes.

Expression of INHBB and USP2 in medulloblastoma

In our evaluation involving mRNA quantification across three distinct human cell lines (HEB, Daoy, and D283), it was observed that the mRNA expression of INHBB and USP2 was comparatively subdued in the Daoy cell line (SHH MB) while being notably pronounced in the D283 cell lines (corresponding to Group 3 medulloblastoma). Subsequent immunohistochemical assessments indicated that the protein expression of INHBB and USP2 was significantly more prominent in the Group 3 medulloblastoma subtype relative to other subtypes. These findings furnish novel perspectives that could be pivotal in the future stratification and differentiation of medulloblastoma subtypes.

Expression patterns across molecular subgroups

To validate our computational findings, we analyzed INHBB and USP2 expression across the four molecular subgroups of medulloblastoma. Box plot analysis revealed significantly higher expression of both genes in Group 3 and Group 4 subtypes compared to WNT and SHH subtypes (p < 2.22e-16 for both comparisons (Figure 12(d)). For INHBB, Group 3 showed approximately five-fold higher expression than WNT/SHH subtypes, while USP2 demonstrated even more dramatic differences with over 20-fold higher expression in Group 3/Group 4 compared to WNT/SHH subtypes. These findings were corroborated by our qRT-PCR validation in cell lines, where D283 cells (Group 3) showed markedly elevated expression of both genes compared to HEB and DAOY cells. qRT-PCR analysis revealed significantly higher mRNA expression of INHBB and USP2 in D283 (Group 3) compared to Daoy (SHH) and HEB cells (Figure 12(a)). Immunohistochemistry confirmed elevated protein levels of both markers in Group 3 tumors (Figure 12(b)–(d)), supporting their subtype-specific prognostic relevance.

Figure 12.

Expression of INHBB and USP2 in medulloblastoma. (a) mRNA expression of INHBB and USP2 in three human cell lines. (b) IHC scoring results of INHBB and USP2 in medulloblastoma patients. (c) Expression of INHBB and USP2 in different subtypes of medulloblastoma. (d) Expression of INHBB and USP2 across four molecular subgroups. Box plots show significantly higher expression in Group 3 and Group 4 compared to WNT and SHH subtypes (Wilcoxon test, p < 2.22e-16).

INHBB: inhibin beta B; USP2: ubiquitin-specific peptidase 2.

Discussion

Medulloblastoma, predominantly situated in the posterior fossa and originating from the vermis of the cerebellum, constitutes 60% of embryonal neoplasms within the pediatric intracranial domain. ¹ Despite advances in multimodal therapeutic strategies encompassing surgical intervention, radiotherapy, and adjuvant chemotherapy, ~30% of patients fail to achieve complete remission, with challenges such as postoperative recurrence and treatment-related toxicities significantly impinging upon patient prognosis. ⁵ Challenges, such as postoperative recurrence and the deleterious effects of radiotherapy and chemotherapy, significantly impinge upon patient prognosis. The integration of molecular subtyping into the World Health Organization’s classification for CNS pathologies has revolutionized medulloblastoma diagnostics, yet the need for more precise prognostic stratification within established molecular subgroups remains critical for optimizing therapeutic strategies.

Prior investigations have elucidated that DNAm, as opposed to traditional genomic analysis, exerts a more profound influence on the initiation and recurrence of neoplastic conditions. ¹⁰ DNAm, prevalent in the ontogenesis of tumors, harbors pertinent oncogenic signals and mirrors the aberrant regulation of gene expression. Consequently, DNAm status can be contemplated as a surrogate indicator of gene expression dynamics. In this study, ROC curves, derived from the prognostic attributes of 25 model genes, substantiated the model’s robustness. Furthermore, we devised a nomogram to graphically represent and forecast the survival probabilities of patients at 3- and 5-year intervals. our model offers additional prognostic granularity beyond traditional molecular subgrouping, successfully identifying high-risk patients within traditionally “favorable” subgroups and low-risk patients within “unfavorable” subgroups. This enhanced stratification capability has profound implications for treatment optimization, potentially allowing for treatment intensification in high-risk patients while enabling de-escalation strategies for low-risk cases.

Our identification of INHBB and USP2 as key prognostic markers represents a significant advancement in understanding Group 3/Group 4 medulloblastoma biology. INHBB, a member of the TGF-β superfamily, plays crucial roles in regulating cell proliferation, differentiation, and apoptosis. The consistently elevated expression of INHBB in Group 3/Group 4 medulloblastomas compared to WNT/SHH subtypes (with approximately five-fold higher expression, p < 2.22e-16) suggests its fundamental involvement in the aggressive phenotype characteristic of these high-risk tumors. The TGF-β signaling pathway has been extensively implicated in promoting tumor progression through multiple mechanisms, including immune suppression, extracellular matrix remodeling, and epithelial-mesenchymal transition. In the context of medulloblastoma, elevated INHBB expression may contribute to treatment resistance and metastatic potential, making it an attractive target for therapeutic intervention.

USP2, a deubiquitinating enzyme that regulates protein stability, demonstrated even more dramatic differential expression patterns, with over 20-fold higher expression in Group 3/Group 4 compared to WNT/SHH subtypes. The significant negative correlation between USP2 and MYCN expression (r = −0.566) observed in our analysis reveals complex regulatory networks that may be central to medulloblastoma pathogenesis. This relationship is particularly intriguing given MYCN’s established role in medulloblastoma, especially in SHH subtypes where gene amplification is common. USP2 may modulate MYCN protein stability or activity through direct deubiquitination or indirect regulatory mechanisms, potentially contributing to the distinct molecular characteristics and clinical behaviors observed across different medulloblastoma subtypes. However, there is currently no direct experimental evidence for a relationship between USP2 and MYCN in medulloblastomas. Based on existing studies of deubiquitinase–E3 ligase regulation, MYCN regulatory mechanisms, and medulloblastoma subtype heterogeneity, we consider this a biologically plausible yet unconfirmed hypothesis. USP2 may indirectly reduce MYCN protein levels by stabilizing E3 ligases such as HUWE1, which promote N-MYC polyubiquitination and degradation.^11,12 Isoform- and tissue-specific functions of USP2, as well as feedback through pathways such as KRAS and AKT/β-catenin, could, under certain contexts, produce this negative correlation.^13–15 This molecular braking mechanism may partially contribute to the complex interplay between USP2 and MYCN. While USP2 negatively correlates with MYCN expression, and WNT-activated tumors (where MYCN can be expressed) are known to have a favorable prognosis, our current data do not support a direct conclusion that high MYCN expression itself is a favorable prognostic marker. Instead, we propose that USP2 may modulate MYCN activity in certain molecular contexts, potentially influencing tumor aggressiveness and therapeutic sensitivity. Further studies will be required to validate this hypothesis.

The robust validation of INHBB and USP2 expression patterns through both computational analysis and experimental validation using qRT-PCR and immunohistochemistry strengthens their potential utility as biomarkers for molecular subtype classification and prognostic stratification. The consistency between mRNA and protein expression levels suggests that these genes may serve as reliable diagnostic markers that could be implemented in clinical practice through routine immunohistochemical staining.

Our findings regarding GLI2 serve as a high-risk gene (coefficient > 0), which aligns with the observation that high expression correlates with poor prognosis. While GLI2 is established as an oncogenic driver in SHH-activated medulloblastomas, its lower expression in higher-risk Group 3/Group 4 tumors suggests fundamentally different regulatory mechanisms operating in non-SHH medulloblastomas. In the absence of active SHH signaling, GLI2 may function through alternative pathways or even exhibit tumor-suppressive roles, emphasizing the molecular heterogeneity of medulloblastoma subtypes and the critical importance of context-specific analysis in cancer research.

Similarly, the observation that MYCN expression appears higher in low-risk groups requires careful interpretation within the framework of molecular mechanisms. This finding underscores the crucial distinction between gene expression levels and gene amplification, which represent different molecular events with potentially divergent prognostic implications. While MYCN amplification is a well-established negative prognostic factor in SHH-activated medulloblastomas, expression levels can be discordant with copy number alterations due to various post-transcriptional regulatory mechanisms, including epigenetic modifications, microRNA regulation, and tumor microenvironment influences. Our analysis measured mRNA expression rather than gene copy number, and the higher expression in our low-risk group (primarily consisting of WNT/SHH patients) may reflect different regulatory contexts where MYCN expression does not correlate with the aggressive phenotype typically associated with its amplification.

While the molecular subtyping of medulloblastoma has provided insights into its onset and development, there is still limited research on the role of the tumor microenvironment in medulloblastoma. The immune microenvironment of medulloblastoma consists of various cell components, including immune and non-immune cells. To explore the correlation between macrophages and prognosis, we found that the content of B cells, inactive macrophages, and other immune cells was significantly elevated in the high-risk group, while the content of macrophages M2 was significantly increased in the low-risk group. The polarization of macrophages is crucial in the progression of intracranial tumors. In the process of medulloblastoma development, the increase in macrophage M2 may be related to the increase in the neuroinflammatory cytokine CCL2. The high expression of CCL2 in the tumor microenvironment promotes the polarization of macrophages M2. ¹⁶ A study involving RNA sequencing and methylation analysis of 70 samples of medulloblastoma in Taiwan, China, revealed ¹⁷ that the SHH subtype of medulloblastoma exhibited high infiltration of macrophages M2. Additionally, the enrichment of macrophages M2 and the expression of their associated genes were associated with favorable outcomes for patients. Therefore, the expression of macrophage M2 and their associated genes may serve as indicators of good prognosis for SHH subtype medulloblastoma. Tumor-associated macrophages (TAMs), as important immune cells in the tumor microenvironment, ¹⁸ have been a subject of controversy in terms of prognosis in different medulloblastoma subtypes. TAMs have a distinct composition in different subtypes of medulloblastoma. Compared to other subgroups, the SHH subtype of medulloblastoma has a higher content of TAMs, which may be related to the expression of monocyte chemoattractant protein-1. ¹⁶ A decrease in TAM content inhibits tumor recurrence and metastasis, while an increase in macrophage M1 content is associated with poorer prognosis in SHH subtype medulloblastoma patients. ¹⁹ Macrophages M1 and M2 exhibit a certain degree of plasticity, and their conversion may have significance in medulloblastoma treatment. Currently, there are limited immunotherapies available for medulloblastoma, and studies have shown that the content of TAMs can impact the effectiveness of immunotherapy. Particularly in patients for whom conventional radiotherapy and chemotherapy are ineffective, TAMs can limit the efficacy of immune checkpoint blockade, ²⁰ indicating that TAMs may be one of the novel targets for future tumor treatment.

Our analysis of immune regulatory factors revealed significant correlations with risk scores, particularly for CXCL12 (r = 0.315, p = 0.004) and its receptor CXCR4 (r = −0.309, p = 2.4e-08). The CXCL12/CXCR4 axis represents a critical pathway for immune cell trafficking and has been implicated in treatment resistance across multiple cancer types. CXCL12 secreted by monocytes facilitates their differentiation into various macrophage subtypes through CXCR4 and CXCR7 receptor engagement. Therapeutic targeting of this pathway, which has shown promise in leukemia treatment, may offer new opportunities for medulloblastoma immunotherapy. ²¹ The correlation patterns observed for immune checkpoint molecules, including CTLA4 (r = 0.122, p = 5e-04) and PDCDLG2 (r = −0.176, p = 4.7e-06), suggest that our risk model captures important aspects of immune dysregulation in medulloblastoma. These findings support the potential for immune checkpoint inhibition in selected patients, particularly those in high-risk categories where conventional therapies have limited efficacy.

In this study, we observed high expression of INHBB and USP2 in the D283 medulloblastoma cell line (Group 3 subtype). Subsequent investigations revealed a significant negative correlation between the expression of USP2 and MYCN, which is one of the commonly implicated oncogenes in cancer. In medulloblastoma, the activation of certain mitotic signaling pathways, such as the WNT and SHH pathways, can contribute to tumor development. MYCN plays a critical role in regulating both normal and aberrant cerebellar development. In SHH subtype medulloblastoma, MYCN gene amplification is relatively common, particularly on chromosome 2, which includes the MYCH 3 gene. ³ This implies a high dependency on MYCN in the formation of the SHH pathway. Moreover, elevated expression levels of MYC protein are a common feature in the majority of Group 3 subtype medulloblastomas. Our analysis identified MYC as a crucial target in Group 3 subtype medulloblastoma. The expression of MYNC was found to be lower in the high-risk group, suggesting potential heterogeneity and evolutionary processes within medulloblastoma. This could be attributed to variations in tumor development stages or molecular events, leading to differential MYC protein levels. In addition, the immune environment of the tumor may influence the expression patterns of tumor cells, potentially indicating immune escape mechanisms or an immunosuppressive microenvironment in high-risk tumors resulting in decreased MYC protein expression. While MYCN is an attractive therapeutic target, direct targeting of MYCN is challenging due to its inherently disordered protein structure. ²²

Our research identified differences in pathways such as autophagy and MTORC1_SIGNALING between different risk groups. Targeting autophagy represents a potential approach in cancer therapy for inhibiting tumor progression. Through drug analysis, we have found that drugs like embelin and atovaquone have the theoretical potential to reverse tumor progression. In particular, embelin, a quinone compound, has demonstrated promising anti-cancer activity by targeting multiple signaling pathways, including mTOR, AKT, and STAT3, thus fostering cellular autophagy and apoptosis. ²³ Atovaquone, a mitochondrial inhibitor, can partially reverse or delay tumor progression. Recent studies have shown that localized injection of mitochondrial-targeted atovaquone into primary tumors in mouse experiments can trigger effective T cell immune responses, affecting both local and distant tumor sites. ²⁴ Atovaquone has also exhibited some antitumor activity in vitro. The mTOR pathway, pivotal in regulating growth and homeostasis, forms two discrete multiprotein complexes: mTORC1 and mTORC2. In cerebellar development, the expansion of cerebellar granule neuron precursors requires the support of SHH and IGF signaling pathways. ²⁵ IGF-II, as a downstream transcriptional target of SHH, activates the IGF pathway, leading to mTORC1-mediated inhibition of 4E-BP1 phosphorylation, thus promoting protein synthesis and contributing to the development of SHH subtype medulloblastoma. ²⁶ In our study, we found that the second-generation mTOR inhibitor NVP.BEZ235 exhibited sensitivity to medulloblastoma-related prognosis risks. Cisplatin, a commonly used drug in medulloblastoma chemotherapy, faces resistance in some patients. BEZ235, as a PI3K/mTOR pathway-targeted inhibitor, can alter autophagic function induced by cisplatin in in vitro experiments, increasing cisplatin sensitivity. ²⁷ Particularly, in glioma patients, the combination of BEZ235 and temozolomide can reduce PI3K/mTOR pathway activity and inhibit tumor proliferation. ²⁸ Targeting the PI3K/mTOR pathway in medulloblastoma has become an interest among researchers. Through combination therapy, it may be possible to improve drug resistance in medulloblastoma treatment.

Several important limitations must be acknowledged in interpreting our findings. First, our analysis was constrained by the gene content of publicly available datasets, resulting in the exclusion of potentially important genes such as CMYC from the expression analysis due to its absence in the GSE85217 array platform. This limitation highlights the challenges of retrospective analyses using existing datasets and emphasizes the need for comprehensive genomic profiling in future prospective studies.

While SLFN11 has demonstrated prognostic significance in medulloblastoma in previous studies, it was excluded from our integrated model due to insufficient differential methylation between molecular subgroups according to our stringent criteria (adj. p < 0.05 and |logFC| > 0.5). This exclusion underscores the trade-offs inherent in integrated analysis approaches, where genes must demonstrate consistent patterns across multiple data types to be included in the final model.

Third, our computational drug sensitivity predictions, while providing valuable therapeutic insights, require extensive experimental validation using appropriate cell line models and in vivo studies to confirm clinical relevance. The transition from computational predictions to clinical applications represents a significant challenge that will require substantial additional research investment.

The retrospective nature of this analysis using publicly available datasets limits the generalizability of findings and introduces potential selection biases. Prospective validation in independent cohorts representing diverse patient populations and treatment approaches will be essential for confirming the clinical utility of our findings. In addition, the absence of detailed treatment information in the public datasets prevents assessment of treatment-specific biomarker relationships, which could be crucial for optimizing therapeutic applications.

Our study design did not include prospective sample size calculations, as we utilized all available data from public repositories. While this approach maximizes statistical power, it limits our ability to make definitive statements about the minimum sample sizes required for clinical implementation of our findings. Future validation studies should incorporate appropriate power calculations to ensure robust clinical translation.

This comprehensive integrated analysis represents a significant advancement in our understanding of medulloblastoma molecular heterogeneity and provides a foundation for improved patient care. The identification of INHBB and USP2 as novel prognostic biomarkers expands the repertoire of clinically relevant molecular markers and offers new targets for therapeutic development. The enhanced prognostic stratification provided by our 25-gene model has immediate implications for clinical trial design, patient counseling, and treatment selection.

The insights into immune microenvironment differences between risk groups provide important groundwork for immunotherapy development in medulloblastoma, a field that has lagged behind adult brain tumors and other pediatric cancers. The identification of specific immune regulatory pathways associated with prognosis offers targets for combination immunotherapy approaches that could significantly improve outcomes for high-risk patients.

Furthermore, our findings contribute to the growing body of evidence supporting the clinical utility of integrated multi-omics approaches in cancer research. The demonstration that methylation–transcriptome integration provides superior prognostic information compared to single-platform analyses reinforces the value of comprehensive molecular profiling, and supports continued investment in multi-omics research infrastructure.

Conclusion

This comprehensive integrated analysis of methylation and transcriptome data has successfully identified novel prognostic biomarkers for medulloblastoma. The 25-gene risk model demonstrates robust predictive capability, offering enhanced prognostic stratification beyond traditional molecular subgrouping. The identification of INHBB and USP2 as key prognostic genes provides new insights into the molecular mechanisms underlying medulloblastoma progression, particularly in high-risk Group 3 and Group 4 subtypes. These findings have important implications for personalized treatment strategies and may guide the development of targeted therapeutic approaches. The distinct immune microenvironment profiles associated with different risk groups further suggest potential for immunotherapy optimization. Future prospective validation studies and experimental investigations of the identified biomarkers will be essential for translating these findings into clinical practice and improving outcomes for medulloblastoma patients.