Integrated Molecular Profiling Improves Subtype Classification and Reveals Inherited Susceptibility in Medulloblastoma: Insights From a Real‐World Cohort

ABSTRACT

Background

Medulloblastoma (MB) is a heterogeneous pediatric brain tumor characterized by distinct molecular subtypes. Although genomics and transcriptomics have improved subtype classification and informed targeted therapies, the clinical utility of integrated molecular profiling in real‐world settings remains incompletely defined.

Methods

We conducted a retrospective analysis of 131 patients with MB from a Chinese cohort. All cases underwent both targeted genomic sequencing and NanoString‐based transcriptomic profiling. We evaluated the impact of integrated molecular profiling on diagnostic refinement, identification of germline predisposition, and detection of actionable alterations.

Results

Incorporation of genomic data led to optimized or revised molecular classification in 67.2% of patients. Germline variants in Fanconi anemia pathway genes (FANCL, FANCC, NBN) were significantly enriched among patients with Group 3 and Group 4 MB, suggesting a potentially inherited susceptibility component. Furthermore, 52.7% of patients harbored potentially actionable somatic mutations, supporting the feasibility of precision‐targeted therapeutic approaches. Notably, we identified one case of WNT‐subtype MB harboring a homozygous deletion of CDKN2A, which was associated with unusually poor clinical outcome.

Conclusions

Integrated genomic and transcriptomic profiling substantially improves molecular subtyping of MB and reveals inherited risk factors relevant to specific subgroups. Our findings support the implementation of combined molecular diagnostics in routine clinical management of MB and underscore their potential in guiding individualized treatment strategies.

1. Introduction

Medulloblastoma (MB), a cerebellar tumor, has an annual age‐adjusted incidence of 2.0–5.8 cases per million globally [1]. It is a leading cause of cancer‐related mortality in children under 16 years of age and ranks among the most common malignant central nervous system (CNS) tumors in pediatric patients [2]. Over the past two decades, the biology of MB has undergone intense scrutiny through comprehensive genome‐wide, transcriptome, and methylome analyses [3, 4]. The 2021 World Health Organization (WHO) recommends the use of Nanostring analysis and DNA methylation profiling for the MBs, molecularly defined [5]. Although IHC techniques are widely implemented by major centers in China for the diagnosis and subtyping of MB, the Nanostring assay has emerged as a promising solution to overcome the limitations of immunohistochemistry (IHC), offering higher sensitivity and accuracy, coupled with advantages in terms of cost‐effectiveness and turnaround time [6]. Recent advancements in genomics, gene expression profiles and DNA methylation have identified over 12 distinct subtypes within the previously established four subgroups of MB [7]. These developments improve molecular classification and enhance risk stratification. For instance, in SHH‐associated MB, TP53 mutations are associated with extremely poor prognosis [8], while amplification of the MYC or MYCN oncogenes is closely linked to inferior survival outcomes and associated with various other high‐risk features [9]. In spite of advancements in our understanding of MB biology, treatment outcomes remain largely unchanged [10].

Despite these advancements in the diagnosis and stratification of MB, several issues remain unresolved in the clinical context, including the differentiation diagnosis, disease progression analysis, and the development of effective treatment strategies. Therefore, the combination of genomics and Nanostring analysis holds significant potential for exploration in the diagnosis and treatment of MB.

This study encompassed a cohort of 131 MB patients from China, utilizing a combination of expression profiling and genomic analysis to investigate the implications of integrated genomic data in the diagnosis and management of MB. Our findings indicate that the incorporation of genomic data facilitated an optimized diagnosis in 67.2% of patients. Furthermore, we identified that carriers of germline variants in the Fanconi anemia (FA) gene are at an elevated risk for developing MB, particularly among those with Group 3 and Group 4 subtypes. Additionally, we identified that 52.7% of the patients harbored actionable mutations amenable to targeted therapeutic strategies. These findings underscore the critical role of combined genomic analysis in advancing personalized medicine for the treatment of MB.

2. Patients and Methods

2.1. Patients

In the retrospective study conducted from 2020 to 2023, a total of 557 patients pathologically confirmed as having MB were enrolled for molecular subtyping. Among them, 131 patients with available tumor tissues and matched blood samples were collected for the detection of genomic variants. Clinical data, including patient age at diagnosis, gender, and primary tumor location at initial presentation, were reviewed. Tumor location was defined based on preoperative neuroimaging (magnetic resonance imaging) and corroborated by surgical and pathological records when applicable; secondary dissemination or metastatic involvement was not included. The study protocol was approved by the Institutional Review Board of Nanjing Simcere Medical Laboratory Science (NSML‐IRB‐202412‐MS02). and all patients provided written informed consent.

2.2. Sample Collection and DNA/RNA Extraction

Formalin‐fixed paraffin‐embedded (FFPE) tumor blocks or frozen fresh tumor tissues were extracted using the gDNA Tissue Extraction Kit (Concert). RNA was extracted from FFPE tissue using the RNeasy FFPE kit (Qiagen, Valencia, CA). Matched peripheral blood gDNA was extracted using the magnetic universal gDNA kit (TIANGEN). The quantification of the extracted gDNA and RNA was measured using the Qubit dsDNA HS (high sensitivity) assay kit on the Thermo Fisher Scientific Qubit Fluorometer and Agilent 4200 TapeStation (Agilent).

2.3. Nanostring nCounter Element Analysis

Molecular subtyping of MB was conducted using a custom Nanostring nCounter elements panel based on gene expression profiling. The CodeSet comprised 22 different genes, with specific genes assigned to each subtype: WNT (WIF1, TNC, GAD1, DKK2, and EMX2), SHH (PDLIM3, EYA1, HHIP, ATOH1, and SFRP1), Group 3 (IMPG2, GABRA5, EGFL11, NRL, MAB21L2, and NPR3), and Group 4 (KCNA1, EOMES, KHDRBS2, RBM24, UNC5D, and OAS1). Probes in the CodeSet were provided by Nanostring Technologies. Sample preparation, hybridization, detection, normalization, scanning, and analysis were performed following the instructions provided by Nanostring Technologies (Nanostring Technologies, Seattle, WA). MB molecular subtyping and class prediction were performed as described [11].

2.4. Next‐Generation Sequencing (NGS)

Approximately 200 ng of input gDNA underwent shearing into 200 – 300 bp fragments through enzyme treatment. The KAPA Hyper DNA Library Preparation Kit from Roche Diagnostics facilitated end repair of the sheared DNA, while the VAHTSTM Universal DNA Library Prep Kit for Illumina (Vazyme) enabled A‐tailing and the synthesis of indexed paired‐end adaptors for the SimcereDx Illumina platform. Size selection was performed to eliminate unligated adaptors using Agencourt AMPure XP beads from Beckman Coulter. Subsequently, ligation products were amplified via polymerase chain reaction (PCR) to generate a hybridization prelibrary. The final qualified DNA libraries were prepared and sequenced by the College of American Pathologists (CAP)‐accredited central laboratory at Jiangsu Simcere Diagnostics Co. Ltd. (Nanjing, China), utilizing the Illumina NovaSeq6000 platform, following the manufacturer's instructions for 150 bp paired‐end sequencing.

2.5. Bioinformatics for Genomic Alterations

The sequencing raw data was processed to generate FASTQ files, which underwent quality control using the fastp software (V.2.20.0) to trim adapter sequences and eliminate low‐quality bases. The Burrows‐Wheller Aligner (BWA‐MEM v.0.7.17) algorithm (available at http://bio‐bwa.sourceforge.net/index.shtml) was then employed to align the clean paired‐end reads to the hg19 human genome reference (UCSC hg19/GRCh37). PCR duplicate reads were removed using Dedup with Error Correct. Single‐nucleotide variations (SNVs) and insertion or deletions (Indels) were identified and annotated using VarDict (v.1.5.7) and InterVar, respectively. Subsequently, common single nucleotide polymorphisms (SNPs) present in public databases, including the 1000 Genome Project (August 2015) and Exome Aggregation Consortium Browser 28, were filtered out. Factera (v1.4.4) and CNVkit (dx1.1) were employed for the analysis of fusions and copy number variations (CNVs), respectively.

To focus solely on significant mutations, a minimum allelic mutation frequency threshold of 2% was applied to SNVs, Indels, and fusions. The Variant Calling Format (VCF) was then annotated using ANNOVAR, and variants were cross‐referenced with records in ClinVar, HGMD, or an in‐house database. Subsequently, germline variants were categorized as pathogenic (P), likely pathogenic (LP), uncertain significance, likely benign, or benign following the 2015 ACMG Guidelines. To determine the tumor mutation burden (TMB), nonsynonymous somatic mutations within the coding region were aggregated, excluding alterations listed in COSMIC as known somatic alterations. Homopolymer repeat loci on the panel, with sufficient coverage, were chosen to evaluate microsatellite instability (MSI). Reads successfully mapped to each locus were extracted from the deduplicated BAM file, and an MSI sensor was utilized to assess the stability of each locus. The MSI score was calculated as the percentage of unstable loci, and samples surpassing the cutoff value were classified as MSI‐H.

2.6. Statistical Analysis

Statistical analysis and visualization were conducted using R 4.3.0 (available at http://www.R‐project.org). Descriptive statistics were generated using the R package tableone (version 0.13.2). The chi‐squared test was utilized by default for categorical variables, while analysis of variance (ANOVA) was employed by default for continuous variables. The odds ratio of genes and the risk of MB were calculated via logistic regression. Additional R packages utilized for visualization include ggplot2 (version 3.5.0), ggsci (version 3.0.1), and ComplexHeatmap (version 2.16.0).

3. Results

3.1. Patient Characteristics of the Study Cohorts

A total of 131 cases of histologically suspected MB were included. All 131 patients were classified using Nanostring analysis. In addition, 55 patients underwent whole‐exome sequencing (WES), 78 patients were sequenced using the Onco‐neuro 353 panel, and two patients underwent both WES and the Onco‐neuro 353 panel to obtain molecular data of accurate diagnosis and treatment hints (Figure 1).

FIGURE 1.

Study Scheme of Patient Cohort. Patients diagnosed with medulloblastoma underwent Nanostring assay to identify molecular subtypes, with a total of 131 patients also being analyzed through Next‐Generation Sequencing (NGS) and Whole Exome Sequencing (WES) as part of an integrated dataset. SNV, Single Nucleotide Variant.

The mean age at diagnosis was 10.3 years (range, 0–46 years). Among the patients, 67.2% were diagnosed before the age of 10, and 12.2% occurred in adult (> 18 years). The overall male: female ratio was 1.7:1. Primary tumor location at initial presentation were most commonly the fourth ventricle (48, 36.6%), followed by the cerebellum (35, 26.7%), and the cerebellar vermis (17, 13.0%). Lesions were uncommon in the brainstem (1.5%), cerebellopontine angle (3.1%) and spinal cord (1.5%). In five patients, tumors involved more than one contiguous anatomical location at presentation, as determined by preoperative imaging and operative records. Among the four histological category, the most common was classic MB (103, 78.6%). The least common subtypes were MB with extensive nodularity (MBEN, 2.3%) and large cell/anaplastic MB (LC/A, 2.3%), while the desmoplastic/nodular MB (D/N) accounted for 11.5%. Seven patients (5.3%) were classified as not elsewhere classified (NEC) and needed further differential diagnosis. The clinical and histological features of these patients are summarized in Table 1. Compared with previous studies, the proportions of the D/N MB and LC/A MB in our population was relatively lower, while the proportion of the classic MB and MBENs were similar. Distinct trends in clinical features and molecular alterations were observed across the four molecular subtypes of MB. Tumor location (p < 0.001), histological classification (p = 0.004), sex pdistribution (p = 0.003), and age‐specific incidence (p = 0.122) varied significant across molecular subtypes. Group 3 MB are rarely observed in adults, whereas the SHH‐activated showed the highest prevalence among adult patients. Nearly all WNT‐activated tumors, along with a substantial proportion of Group 3 and Group 4 tumors, were of the classical histological subtype. In our study population, SHH‐activated MB exhibited the greatest histological heterogeneity (Figure 2).

TABLE 1.

The clinical characteristics of patients with MB in our cohort.

Characteristics	N (% of total)
Total number of patients	131
Mean age at diagnosis, years	10.3 (0–46)
0–10	88 (67.2%)
10–20	29 (22.1%)
20–30	6 (4.6%)
30–40	5 (3.8%)
40–50	3 (2.3%)
Gender
Female	48 (36.6%)
Male	83 (63.4%)
Histopathology
Classic	55 (42.0%)
Desmoplastic or nodular	15 (11.5%)
Extensive nodularity	3 (2.3%)
Large cell or anaplastic	3 (2.3%)
None	55 (42.0%)
Location
Supratentorial	1 (0.8%)
Temporal lobe	1
Infratentorial	112 (85.5%)
Cerebellar vermis	17
Cerebellum	35
Fourth ventricle	48
Spinal cord	2
Cerebellum + Cerebellar vermis	2
Fourth ventricle + Cerebellar vermis	1
Brainstem + Cerebellar vermis	1
Brainstem	1
Cerebellopontine angle	4
Cerebellum + Spinal cord	1
None	18 (13.7%)
Medulloblastoma subtype
WNT	18 (13.7%)
SHH	42 (32.1%)
Group 3	19 (14.5%)
Group 4	52 (39.7%)

FIGURE 2.

Graphical Summary of the medulloblastoma. Composition and clinical features of MB subtypes determined by the NanoString assay, including age, gender, location, histopathology, germline P/LP variants, and CNV burden. MB medulloblastoma. P/LP pathogenic or likely pathogenic. CNV, copy number variations.

3.2. The Mutational Landscape of MB Patients

The Nanostring assay data, prevalent CNVs, mutations, and affected pathways of 131 MB patients are summarized in Figure 3a. In our cohort, frequently observed somatic variants overlapped with known MB driver genes, including DDX3X, PTCH1, CTNNB1, KMD6A, KMT2D, and others, with distinct clustering observed across molecular subtypes (Figure 3b).

FIGURE 3.

Mutational signatures in MB. (a) Mutational signatures of MB stratified by molecular subtype. Mutated genes are ordered by the mutation frequencies. (b) Four molecular types of high‐frequency mutation genes.

We evaluated two frequently altered pathways in MB. WNT‐activated and SHH‐activated MB frequently exhibited concurrent aberrations involving both pathways. Mutations in the CTNNB1 were exclusively observed in the WNT subtype. PTCH1 mutations were primarily observed in the SHH subtype, with exceptions noted in two patients classified as WNT subtype. Notably, TP53 mutations occurred more frequently in the WNT subtype than in the SHH subtype. TP53 mutation and alterations in the SHH pathways, although relevant in SHH‐activated MB, were not exclusive to the subtype.

The five most frequent CNVs identified were CDK6 amplification (14.5%), MET amplification (9.2%), MYC amplification (8.4%), ERBB2 amplification (6.1%), and CDK4 amplification (5.3%). Chromosome 17q (53.4%) was the most frequently altered region, followed by17p (51.1%), 7p (41.2%), 16p (40.5%), 8p, and 8q (36.6%). Loss of chromosome 6 was a more specific feature of the WNT subtype (88.9%).

Consistent with previous reports, Group 3 MB demonstrated a higher CNV burden than the SHH and WNT subtypes (p = 0.049 and 0.011, respectively) (Figure 2). Our results showed that the CNV burden in the SHH subtype was not significantly different from that observed in Group 4.

A total of 15 P/LP germline variants across 109 cancer susceptibility genes were identified in 14 patients (10.7%) according to the 2015 American College of Medical Genetics and Genomics (ACMG) guideline. Two patients were diagnosed with Gorlin syndrome (SUFU), three with Li‐Fraumeni syndrome (TP53), and one with Lynch syndrome. Additionally, seven patients showed P/LP mutations associated with FA (NBN, FANCL, FANCA, SLX4, FANCC, FANCD2, and BRCA2) (Supplementary Table S1). Among the four subtypes, three cases (16.7%) of the WNT type, five cases (11.9%) of the SHH type, and four cases of Group 4 (7.7%), one case (5.3%) of Group 3 exhibited P/LP mutations (Figure 2).

3.3. Integrated Genomics Analysis Refine Histopathological Diagnoses and Risk Stratification

Molecular classification was performed using transcriptomic data from the Nanostring assay. The proportions of the four molecular subgroups, WNT‐activated, SHH‐activated, Group 3, and Group 4 were 13.7%, 32.1%, 14.5%, and 39.7%, respectively. Classic MB is observed across all four molecularly defined subtypes and was most prevalent in Group 4 and SHH‐activated MB. D/N MB and MBEN are classified primarily as SHH‐activated, and, less frequently, as Group 4, with the SHH‐activated representing the predominant subtype (66.7%), whereas Group 4 was rarely reported in previous studies [12]. Most LC/A MB belonged to either the SHH‐activated or Group 3 subtype. All seven NEC patients were reclassified into molecular subtypes. Molecular classification using the Nanostring assay significantly refined MB subtypes assignments relative to the original histopathological diagnoses. When genomic profiling was integrated with Nanostring‐based transcriptomic classification, 67.2% of cases underwent refinement of their integrated diagnostic interpretation, including clarification of histologically ambiguous tumors and risk stratification refinement within established molecular subgroups, rather than molecular subgroup reassignment alone. For example, one tumor initially classified as histologically NEC and assigned to Group 4 by Nanostring analysis was reclassified as IDH‐mutant astrocytoma based on integrated genomic findings (Figure 4a).

FIGURE 4.

Integrative genomic analyses for diagnosis and therapeutic decision‐making. (a) Sankey diagram illustrating the relationship between initial histological diagnosis, Nanostring‐based molecular subgrouping, and integrated diagnostic interpretation after incorporation of genomic profiling, including diagnostic clarification and risk stratification refinement within molecular subgroups. (b) Summary of targeted therapy targets for MB. Nightingale rose chart illustrating the distribution of potential targeted therapy biomarkers. (c) Stacked bar graph of the percentages of COSMIC single nucleotide mutational signatures in FA and non‐FA groups. FA, Fanconi anemia.

3.4. Integrated Genomic Analysis Identifies Germline Risk Genes for MB

We identified seven patients with P/LP variants in the FA‐related genes, including NBN, FANCL, FANCA, SLX4, FANCC, FANCD2, and BRCA2 (Supplementary Table S1). When comparing mutation frequencies of FA genes in our cohort with those in the noncancer gnomAD v.2.1 population [13, 14]. Loss‐of‐function (LOF) variants in NBN, FANCL and FANCC were more common in our cohort when compared to controls (p = 0.021, 0.006, and 0.032, respectively). We further observed that LOF variants in XPA (p = 0.046) and SDHB (p < 0.001) were more prevalent in our cohort. Heterozygous germline variants in NBN, FANCL, FANCC, XPA, and SDHB may increase susceptibility to MB (Table 2).

TABLE 2.

Comparison of the frequency of LoF gene variants between patients with MB and the general population.

Pathway	Gene	Noncancer gnomAD
		MN	SN	LOF MN	LOF SN	OR	95% CI	p
HRR	NBN	1	131	88	117,819	10.29	1.423 – 74.429	0.021
	FANCL	1	131	57	118,202	15.94	2.191 – 116.005	0.006
	FANCA	1	131	289	116,050	3.08	0.429 – 22.114	0.263
	SLX4	1	131	127	73,530	4.45	0.617 – 32.046	0.139
	FANCC	1	131	104	117,819	8.71	1.206 – 62.863	0.032
	FANCD2	1	131	171	117,903	5.30	0.736 – 38.098	0.098
	BRCA2	1	131	352	116,200	2.53	0.353 – 18.158	0.355
Other	XPA	1	131	121	117,940	7.49	1.039 – 54.008	0.046
	SDHB	1	131	18	117,767	50.32	6.669 – 379.710	< 0.001

Abbreviations: LoF, loss of function; MN, mutation number; SN, sumary number.

3.5. CDKN2A/CDKN2B Homozygous Deletion in Rapidly Progression WNT‐Activated MB

Despite the generally favorable prognosis of WNT‐activated MB, with 5‐year progression‐free survival (PFS) and overall survival (OS) rates approaching 100%, we describe a unique case involving an 11‐year‐old patient diagnosed with WNT‐activated MB in the fourth ventricle. This patient experienced disease recurrence only 3 months after surgery and chemotherapy. Genomic analysis revealed a hotspot CTNNB1 mutation and unexpectedly identified a homozygous deletion of CDKN2A/B. At present, the clinical implications of CDKN2A homozygous deletion in WNT‐activated MB remain largely unexplored, suggesting that it may represent a potential risk factor for disease recurrence.

3.6. Targeted Molecular Alterations and Therapeutic Implications, Particularly in Group 3 and Group 4 MB

NGS assays represent a promising avenue for targeted therapy in MB, a condition primarily managed with surgery, radiotherapy, and chemotherapy. As clinical trials become increasingly prevalent, our study identified that 52.7% (69 out of 131) of patients with MB harbored at least one molecular biomarker potentially suitable for precision therapy. Among these, the five most frequently identified biomarkers involved CDK6, MET, PTCH1, ERBB2, and CDK4 (Figure 4b). Among the 14 patients identified with germline mutations, all those classified as Group 3 or Group 4 were carriers of FA‐related genes. Therefore, we investigated the potential influence of FA germline mutations on mutagenesis in these patients. Analysis of mutational signatures can elucidate the contributions of specific genes, such as BRCA1/2 and mismatch repair (MMR) genes, to tumorigenesis. We observed that COSMIC mutational signature 3, which is associated with BRCA mutations, contributed to 10.7% of SNVs in the FA group, compared with 5.7% in the non‐FA group. In contrast, signatures associated with MMR defects (signatures 20 and 26) accounted for 19.0% of SNVs in the FA group, compared with 22.2% in the non‐FA group (Figure 4c). Nonetheless, the mutation counts detected in each specimen by cancer gene panel sequencing were insufficient for robust mutational signature analysis. These findings suggest that patients with Group 3 and Group 4 MB may potentially benefit from treatment with PARP inhibitors.

4. Discussion

In this study, we explored the incorporation of genomic analysis into the NanoString classification of MB. Our results offer significant advancements in the molecular diagnosis and treatment of this complex malignancy, potentially guiding future therapeutic strategies. In terms of population characteristics, the proportions of MBENs and classic MB were consistent with previous studies. However, the proportions of D/N and LC/A MB observed in our study were only about half of those reported in another study [15]. A slightly higher proportion of D/N and MBEN subtypes has been classified as Group 4 MB in prior reports [12]. This difference likely reflects variability in histopathological interpretation across populations in real‐world diagnostic settings. In principle, D/N and MBEN histologies are strongly enriched in SHH‐activated MB. Nevertheless, accumulating molecular evidence indicates that this association is not absolute. Large‐scale molecular studies have demonstrated that a small proportion of tumors with D/N or MBEN morphology can be classified as Group 4 based on transcriptomic profiling. In addition, expression‐based analyses have shown that Group 3 and Group 4 MBs form a biologically meaningful continuum, providing a developmental framework in which such histologic–molecular discordance may occur [16, 17]. In our cohort, a modestly higher proportion of D/N and MBEN tumors was assigned to Group 4. Although this proportion exceeds that reported in reviewed cohorts, it should be interpreted cautiously given the limited number of DN/MBEN cases and the absence of histopathological review. To clarify these histologic‐molecularly discordant cases, we summarized the DN/MBEN tumors assigned to Group 4 in Table 3, including their transcriptomic classification and key genomic features.

TABLE 3.

Summary of desmoplastic/nodular and MBEN medulloblastomas assigned to Group 4 by integrated molecular profiling.

Case ID	Clinicopathology	Nanostring molecular subtype	Key genomic features	SHH pathway alterations	Integrated diagnostic interpretation
MBP_028	desmoplastic/nodular	Group 4	chr17q gain	None detected	MB, Group 4—low risk
MBP_079	desmoplastic/nodular	Group 4	chr 16p partial‐arm loss	None detected	MB, Group 4—low risk
MBP_085	desmoplastic/nodular	Group 4	chr7 gain, chr17 partial‐arm gain, chr11 partial‐arm loss	None detected	MB, Group 4—low risk
MBP_087	extensive nodularity	Group 4	chr9 gain	None detected	MB, Group 4—standard risk
MBP_112	desmoplastic/nodular	Group 4	chr17 partial‐arm gain, chr13 partial‐arm gain, chr11 partial‐arm loss	None detected	MB, Group 4—low risk
MBP_118	desmoplastic/nodular	Group 4	chr17 partial‐arm gain	None detected	MB, Group 4—low risk

Importantly, the overall distribution of the four molecular subgroups in our study closely mirrors that observed in previous studies, suggesting that the observed differences likely reflect variability in histopathological interpretation across populations rather than systematic deviations in molecular classification.

Moreover incorporation of genomic analysis allowed further delineated of molecular subtypes and risk stratification in accordance with WHO classification guidelines. Ultimately, over 67% of the molecular interpretations for MB underwent substantial revision. Notably, a 38‐year‐old male patient initially presented with a space‐occupying brainstem lesion and was molecularly classified as Group 4 MB. However, subsequent NGS performed as part of integrated diagnostics revealed several critical genetic alterations, specifically mutations in IDH1 (c.395G>A, p.R132H), ATRX (c.3734C>G, p.S1245*), and TP53 (c.488A>G, p.Y163C), as well as a homozygous deletion of CDKN2A/B. Following multidisciplinary pathological review, the tumor was subsequently reclassified as an IDH‐mutant astrocytoma, WHO Grade 4. This case illustrates the complexities inherent in tumor classification and highlights the necessity for integrating comprehensive genomic analyses into clinical diagnostics.

Our study indicates that the SHH subtype of MB exhibits significant heterogeneity, with a broad age distribution encompassing all four histological variants, as well as neuroendocrine carcinoma (NEC). Genetic analysis indicates overlap in enriched signaling pathways between the SHH subtype and the WNT subtype. Furthermore, the SHH subtype exhibited a high frequency of chromosomal alterations, comparable to that observed in Group 4. A similar trend has been reported in MB with SHH activation mediated by DNA methylation, which can be classified into four distinct temporal molecular subgroups [7]. Currently, clinical applications of biomarkers reflecting heterogeneity in SHH‐activated MB remain limited beyond TP53 and a few genetic susceptibility genes. However, integration of genomic data with NanoString results provides a robust basis for informed diagnostic and therapeutic decision‐making.

Notably, patients with WNT‐activated MB are typically considered low‐risk, prompting exploration of treatment de‐escalation strategies [18]. The significance of CDKN2A homozygous deletions in WNT‐activated MB remains insufficiently explored. Nevertheless, such deletions have been well established as adverse prognostic markers in gliomas [19] and other malignancies [20, 21]. This observation suggests that CDKN2A homozygous deletion may represent a potential risk factor for disease recurrence in this context.

Large population studies have reported that approximately 10% of patients with MB harbor germline mutations, with six genes, APC, BRCA2, PALB2, PTCH1, SUFU, and TP53, accounting for excess damaging variants in about 6% of MB patients [22]. Consistent with this incidence, our study identified functional LoF variants in several homologous recombination repair pathway genes (FANCL, FANCC, and NBN), suggesting increased susceptibility to MB compared with noncancer populations, particularly among non‐WNT/non‐SHH subtypes. FA genes encode proteins that participate in multiple DNA damage repair pathways, particularly those involved in the resolution of interstrand crosslinks that impede DNA replication [23]. Numerous pathogenic variants in FA genes have been associated with an increased risk of multiple cancers. Previous studies have demonstrated that these mutations specifically affect the ubiquitination of the FANCD2/FANCI complex, a key step in the FA DNA repair pathway. In particular, variants in FANCM and XRCC2 have been associated with an increased risk of breast cancer [24], whereas FANCG variants have been linked to head and neck cancer risk [25].

FA genes are classified as DNA damage response (DDR) genes, and their status in patients with FA indicates heightened sensitivity to chemotherapy, often resulting in improved treatment outcomes [26]. However, this increased sensitivity is accompanied by a higher incidence of complications; indeed, more than half of the reported adverse reactions to radiotherapy are linked to FA [27]. However, the current standard of care MB includes a generalized combination of surgical resection, chemotherapy, and craniospinal radiotherapy. Our findings provide important insights that may help guide treatment strategies and inform comprehensive clinical decision‐making.

The NanoString assay and DNA methylation analyses have elucidated molecular subtypes of MB; however, they may overlook potentially actionable variants. The five most frequently identified biomarkers included CDK6, MET, PTCH1, ERBB2, and CDK4 as reported in previous studies. The CDK4/6‐specific inhibitor palbociclib has been preclinically validated as an effective treatment for MB cells [28]. In addition, c‐Met inhibitors have been shown to significantly reduce tumor growth in xenograft mouse models of MB [29]. Furthermore, phase I and phase II clinical trials demonstrated that sonidegib and vismodegib were well tolerated and exhibited antitumor activity in SHH‐driven pediatric and adult MB [30]. The PARP inhibitor Olaparib exhibits radiosensitizing properties in pediatric MB, and its combination with radiotherapy has been shown to improve survivals in MB models [31]. Although most alterations do not currently confer actionable status in MB, these pathways are already therapeutically tractable in other solid tumors and have demonstrated variable but encouraging biological activity in preclinical MB models. Their presence in a meaningful subset of patients raises the possibility that rational drug‐repurposing approaches or molecularly stratified clinical trials could help define targeted interventions for selected populations. Likewise, alterations affecting the FA and homologous recombination pathways further expand the conceptual framework for exploring DNA damage–response‐directed strategies in MB, particularly as these mechanisms are increasingly recognized across pediatric embryonal tumors.

Taken together, our findings suggest that integrated genomic and transcriptomic profiling provides clinically relevant information beyond molecular subgroup assignment in MB. Within established subgroups, selected genomic alterations may contribute to a more nuanced interpretation of clinical risk, based on biological plausibility and consistency with emerging evidence from prior studies.

The limitations of our study are as follows: first, the sample size is relatively small, although it adequately reflects the characteristics of the MB population and disease. Second, this study is retrospective in nature; therefore, prospective validation in a larger cohort and a more diverse patient population is warranted.

In conclusion, the integration of genomic and transcriptomic profiling improves the precision of MB classification in real‐world clinical practice and identifies genomic features that may aid in the interpretation of clinical risk within established molecular subgroups. This strategy is readily reproducible, relying on Nanostring‐based molecular subtyping combined with routinely available targeted sequencing platforms, and enables the integration of precise molecular classification, identification of risk‐relevant genomic features, and systematic detection of potentially targetable pathway alterations within standard molecular pathology laboratories. Together, these findings provide a practical framework for future translational research and hypothesis‐driven clinical trials in MB.

Author Contributions

Conceptualization: Jiwei Song. Methodology: Tiantian Han, Jiwei Song. Formal analysis: Siqi Chen, Jiwei Song. Investigation: Jiwei Song, Tiantian Han, Dongsheng Chen. Writing – original draft preparation: Jiwei Song, Tiantian Han, Siqi Chen. Writing – review and editing: Ziying Liu, Dongsheng Chen. Supervision: Ziying Liu, Dongsheng Chen.

Funding

The authors have nothing to report.

Ethics Statement

All experiments involving human participants and human tissue samples were conducted in accordance with relevant guidelines and regulations. The study protocol was approved by the Institutional Review Board of Nanjing Simcere Medical Laboratory Science (NSML‐IRB‐202412‐MS02).

Consent

Informed consent was obtained from all subjects and/or their legal guardians prior to sample collection.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Clinical information and germline variants of patients with MBs enrolled in our study.

Data Availability Statement

The raw sequencing data are available under restricted access due to data privacy laws. Data are available on request sharing by sending requests to the corresponding author Ziying Liu (liuziying_foshan@163.com), which will need the approval of the institutional ethical committees. Clinical data were not publicly available due to involving patient privacy but can be accessed from the corresponding author upon request for 3 years; individual deidentified patient data will be shared for clinical study analyses. The remaining data are available in the manuscript, Supporting Information, or Source Data file. This paper does not report original code.