ABSTRACT
DNA methylation‐based classification is a key adjunct to histopathology in CNS tumor diagnostics. Although both fresh‐frozen and FFPE tissues are used, the impact of sampling strategy on classifier performance remains insufficiently characterized. We analyzed 179 CNS specimens (162 tumors and 17 controls), including 42 FFPE and 137 fresh‐frozen samples with ≥ 50% tumor cellularity. Fresh‐frozen cases were stratified into histology‐validated and blind sampling groups. Genome‐wide DNA methylation profiling was performed using Illumina EPIC/EPIC v2.0 arrays and classified using the DKFZ Brain Tumor Classifier, with independent validation using the NIH Methylscape platform. Results were categorized as confirmed, refined, changed, or unclassified and correlated with histopathological diagnoses. FFPE samples showed high diagnostic yield with minimal unclassified results (2%). Among fresh‐frozen specimens, histology‐validated samples demonstrated robust classification, whereas blind sampling showed higher discordance and unclassified rates (22%). Discordant results were confined to the blind‐sampling group and were most consistent with sampling bias and variable tumor representation rather than technical limitations. Methylation profiling confirmed or refined diagnoses in most cases and led to clinically relevant reclassification in a subset. Classifier performance appeared to be influenced more by sampling strategy rather than preservation method, with histology‐guided sampling associated with the most reliable diagnostic outcomes.
Keywords: CNS tumors, DKFZ Brain Tumor Classifier, DNA methylation profiling, FFPE tissue, fresh‐frozen tissue, integrated diagnosis, NIH Methylscape Classifier
DNA methylation profiling of CNS tumors highlights the impact of tissue sampling strategy on diagnostic accuracy. Histology‐guided sampling in both FFPE and validated fresh‐frozen tissue was associated with consistent, reliable classifier performance, whereas blind sampling showed higher rates of discordant or unclassified results. These findings indicate that sampling approach, rather than tissue preservation format alone, is an important factor associated with accurate methylation‐based diagnosis, and support histological verification of tumor adequacy prior to molecular profiling.
1. Introduction
Central nervous system (CNS) tumors constitute a highly heterogeneous group of neoplasms, encompassing diverse histological subtypes, molecular alterations, and clinical behaviors [1]. Over the last decade, the diagnostic approach to CNS tumors has undergone a paradigm shift with the integration of molecular parameters into the World Health Organization (WHO) classifications of 2016 and 2021 [1, 2]. Among these, DNA methylation profiling has emerged as a robust and reproducible adjunct to histopathology and immunohistochemistry, offering a more objective and precise framework for tumor classification [3, 4, 5]. The methylation classifier has significantly improved the resolution of diagnostically challenging cases, enabled identification of novel tumor subgroups, and enhanced prognostic and predictive correlations [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13].
In clinical practice, DNA methylation profiling can be performed on both fresh‐frozen and formalin‐fixed paraffin‐embedded (FFPE) tissue [3, 14]. Fresh‐frozen samples are often considered superior due to better DNA integrity and reduced crosslinking artifacts [15, 16]. However, their use is limited by practical challenges such as restricted availability, lack of histological guidance during sampling, and variable tumor content, which may compromise diagnostic accuracy [6, 17, 18, 19]. On the other hand, FFPE tissue, despite potential DNA fragmentation or chemical modification, allows for direct microscopic verification of tumor content prior to extraction, thereby reducing sampling errors [4, 14]. With recent advances in extraction protocols and array technologies, FFPE tissue has become increasingly reliable for methylation analysis [3, 6, 14].
Although the feasibility of DNA methylation profiling from both fresh‐frozen and FFPE samples has been demonstrated, relatively few studies have systematically compared their diagnostic performance in large real‐world cohorts [14, 16, 20]. To address this gap, we analyzed 179 CNS tumor cases including controls in which methylation profiling was attempted on both tissue types, with the objective of evaluating classification success, discrepant results, and practical limitations in routine diagnostic practice.
2. Materials and Methods
2.1. Study Cohort and Histopathological Evaluation
A total of 179 human brain tissue samples were included in this study, comprising 162 tumor specimens and 17 non‐neoplastic controls. Both FFPE blocks (n = 42) and fresh‐frozen tissues (n = 137) were analyzed. Relevant clinical information, including patient age, sex, and histopathological diagnosis, was retrieved from institutional records. The age of patients ranged from 4 months to 75 years, with a male‐to‐female ratio of approximately 1.8:1. Histopathological diagnosis was established according to the WHO Classification of Tumors of the Central Nervous System, 5th edition (WHO CNS 5). All cases underwent detailed histomorphological evaluation supplemented by an extended immunohistochemical panel, including IDH1 R132H, ATRX, p53, H3K27M, H3K27me3, H3G34R/V, and INI1/BRG1 wherever indicated, along with routine molecular and cytogenetic testing, including FISH for 1p/19q codeletion, CDKN2A deletion, EGFR amplification, and other relevant markers as required for integrated diagnosis. Cases in which the available histopathological, immunohistochemical, and molecular findings were insufficient for definitive classification within the WHO CNS 5 framework were designated as NOS. Only samples with tumor cell content > 50% on hematoxylin–eosin (H&E) stained sections were considered eligible for analysis. Among the 137 fresh‐frozen specimens, samples were stratified by tissue weight into 50 small‐to‐medium cases (20–70 mg) and 87 large cases (> 70 mg).
2.2. Histopathological Evaluation and Sampling Strategy
To evaluate the impact of tissue handling and sampling strategy on methylation‐based classification, specimens subjected to DNA methylation profiling were stratified into three groups based on tissue type and method of tumor‐content verification:
Group 1: FFPE (Tumor‐Content Verified).
Group 2: Fresh‐Frozen (Histology‐Validated Sampling).
Group 3: Fresh‐Frozen (Blind Sampling).
In Group 1 (FFPE), tumor‐rich areas were selected on H&E‐stained slides prior to DNA extraction, allowing direct microscopic confirmation of tumor cellularity. In Group 2 (large Fresh‐Frozen specimens), a histology‐guided validation workflow was implemented. Serial sectioning was performed, with H&E staining of the first and eleventh sections to identify tumor‐rich regions and confirm a minimum tumor cellularity of > 50%. DNA was then extracted from the intervening unstained sections. In Group 3 (small/medium fresh‐frozen specimens), tissue availability was limited and histological validation could not be performed. DNA extraction was therefore carried out without prior microscopic verification of tumor content (blind sampling). For all cases, representative sections were independently reviewed by two neuropathologists (V. S. and S. M.) to confirm the histopathological diagnosis and estimate tumor cellularity wherever material was available. Control samples consisted of non‐neoplastic brain parenchyma obtained from epilepsy or trauma‐related resections and were verified to be free of tumor involvement upon neuropathological examination.
2.3. DNA Extraction and Quality Assessment
For FFPE blocks, 5–10 µm thick sections were cut from tumor‐rich regions identified on H&E slides. All FFPE tissues were processed according to standard institutional pathology protocols. Briefly, specimens were fixed in 10% neutral buffered formalin, with fixation durations typically ranging between 12 and 24 h, consistent with routine diagnostic practice. Following fixation, tissues were paraffin‐embedded and stored under standard archival conditions (room temperature, low humidity). Among the 42 FFPE blocks, tissue was archived between 2017 and 2025, corresponding to a median block age of approximately 2 years (range 0–8 years) at the time of methylation analysis; 86% (36/42) dated from the preceding 3 years. Due to the retrospective nature of the cohort, detailed fixation duration records were not uniformly available for all samples; however, all cases underwent histopathological quality assessment prior to molecular analysis. Deparaffinization and DNA isolation were performed using the ReliaPrep FFPE gDNA Miniprep kit (Promega) according to the manufacturer's protocol. For frozen samples, DNA was extracted from 20 to 30 mg of tissue using the Qiagen DNeasy Blood & Tissue kit following standard processing guidelines provided by the manufacturer.
DNA quality was assessed using substrate‐appropriate metrics. FFPE‐derived DNA was evaluated by two complementary measures: ΔCt value from the Illumina Infinium FFPE QC assay, a real‐time PCR assay of DNA amplifiability that determines suitability for the Infinium methylation array (ΔCt < 3 considered array‐suitable) and DNA Integrity Number (DIN; Agilent TapeStation), which quantifies genomic fragmentation. A DIN ≥ 3.0 was applied as a guide, with borderline samples retained when they met the ΔCt criterion. Fresh‐frozen DNA was assessed by concentration and purity using spectrophotometry (A260/A280 ratio) and fluorometric quantification (Qubit dsDNA HS Assay), as DIN and the FFPE‐specific ΔCt assay are not applicable to this substrate. In addition, methylation‐specific quality control was evaluated using established CpG probe detection‐rate thresholds (≥ 96% for fresh‐frozen and ≥ 90%–95% for FFPE‐derived samples); all analyzed samples satisfied these criteria and were therefore deemed suitable for downstream methylation‐based classification of CNS tumors.
2.4. DNA Methylation Profiling
Genome‐wide DNA methylation profiling was performed using the Illumina Infinium Human Methylation BeadChip array (EPIC/EPIC v2.0). Prior to array processing, bisulfite conversion 500–700 ng genomic DNA per sample was carried out using the Zymo EZ DNA Methylation‐Gold Kit following the manufacturer's instructions. Bisulfite‐converted DNA was subsequently processed through the standard Infinium methylation workflow, including whole‐genome amplification, enzymatic DNA fragmentation, BeadChip hybridization, and single‐base extension/staining, in accordance with Illumina protocols. BeadChips were scanned on the Illumina iScan System to generate raw IDAT data used for downstream analysis.
2.5. Classifier Analysis
Methylation‐based classification of CNS tumor was performed using two validated classifiers: the DKFZ Brain Tumor Classifier (version 12.8.1) and the NIH Methylscape Classifier. Classifier predictions were interpreted based on calibrated confidence scores. Predictions with scores ≥ 0.9 are considered “classified.” Scores < 0.9 could still be relevant for cases with low tumor content. Samples scoring below 0.3 were not assigned a definitive methylation class.
2.6. Diagnostic Categorization
Classifier results were systematically compared with the initial histopathological diagnosis and grouped into four categories. Cases were considered confirmed when the diagnosis remained unchanged and was fully supported by methylation profiling. A diagnosis was regarded as refined when it was retained but further subclassification was achieved, such as molecular subgrouping of medulloblastomas. In some instances, methylation profiling provided a completely different assignment, and these were classified as changed or new diagnosis. Finally, cases that failed to achieve a reliable match due to low classifier confidence were labeled as unclassified.
3. Results
A total of 179 samples were subjected to DNA methylation profiling, including 162 tumor samples and 17 non‐neoplastic control brain parenchymal tissues. Of these, 42 samples were FFPE blocks, while 137 were fresh‐frozen tissues (50 small/medium cases and 87 large cases). The study cohort spanned an age range of 4 months to 75 years with a 1.8:1 male‐to‐female distribution. The tumor cohort represented a heterogeneous spectrum of CNS neoplasms. On histological evaluation, the majority of cases were diagnosed as Diffuse High‐Grade Glioma, CNS WHO Grade 4, NOS (n = 90), followed by Meningioma (n = 35), Ependymoma (n = 14), Medulloblastoma (n = 12), Atypical Teratoid/Rhabdoid Tumor (n = 9), and Epithelioid Glioblastoma (n = 2).
Among FFPE samples (n = 42), the majority comprised Diffuse High‐Grade Glioma (n = 40), with one case each of Medulloblastoma and Epithelioid Glioblastoma. Within the histology‐validated frozen cohort (Group 2), cases included Diffuse High‐Grade Glioma (n = 28), Meningioma (n = 22), Ependymoma (n = 9), Medulloblastoma (n = 7), and Atypical Teratoid/Rhabdoid Tumor (n = 4), along with 17 non‐neoplastic control tissues. In the blind‐sampling frozen cohort (Group 3), tumor types included Diffuse High‐Grade Glioma (n = 22), Meningioma (n = 13), Ependymoma (n = 5), Medulloblastoma (n = 4), Atypical Teratoid/Rhabdoid Tumor (n = 5), and one case of Epithelioid Glioblastoma.
When analyzed according to sampling strategy, classifier performance differed across the three groups. In Group 1 (FFPE; Tumor‐Content Verified), successful classification was achieved in 41 of 42 cases (98%). In Group 2 (Fresh‐Frozen; Histology‐Validated Sampling; n = 87), all cases were successfully classified (100%). In contrast, Group 3 (Fresh‐Frozen; Blind Sampling; n = 50) showed 11 classification failures (22%), with all unsatisfactory results within the frozen cohort confined to this subgroup. Importantly, tumor‐type distribution across groups was comparable and did not account for these differences. The unsatisfactory results in Group 3 were most consistent with variable tumor representation within the sampled aliquots, which compromised methylation signal uniformity and classifier confidence. Retrospective histopathological review of corresponding FFPE blocks demonstrated tumor cellularity greater than 50% in all cases; however, the exact aliquots used for DNA extraction were not histologically verified, and local variability in tumor representation cannot be excluded. These findings indicate that, despite adequate DNA quality, biological variability within frozen tissue can significantly influence methylation profiling outcomes. Statistical comparison using Fisher's exact test (two‐tailed), showed no significant association between tissue type and classifier performance (p = 0.2993).
Calibration scores obtained from methylation‐based classification were stratified into three categories: high confidence (> 0.9), intermediate confidence (0.5–0.9), and low confidence (< 0.5) [21]. Among Group 1 FFPE‐derived samples, 64% (27/42) achieved high‐confidence classification with calibration scores > 0.9, 26% (11/42) showed intermediate confidence scores between 0.5 and 0.9, and the remaining 10% (4/42) were classified with low confidence (calibrated scores < 0.5). In Group 2 (Fresh‐Frozen: Histology Validated Sampling), a higher proportion of cases achieved high‐confidence classification, with 77% (67/87) demonstrating calibration scores > 0.9. An additional 14% (12/87) fell within the intermediate confidence range (0.5–0.9), while 9% (8/87) showed low‐confidence scores (< 0.5). In Group 3 (Fresh‐Frozen: Blind Sampling), 68% (34/50) of cases achieved high‐confidence classification (> 0.9), 12% (6/50) demonstrated intermediate confidence (0.5–0.9), and 20% (10/50) were categorized as low confidence (< 0.5).
DNA quality did not account for these differences in classifier performance. For the FFPE cohort (n = 42), DIN had a median of 3.8 (range 2.0–6.4), and ΔCt values from the Illumina Infinium FFPE QC assay had a median of 0.4 (range −2.1 to 2.3), with all 42 samples below the array‐suitability threshold of ΔCt < 3 (Supporting Information S1: Table 1). Neither DIN (Spearman's ρ = −0.27, p = 0.086) nor ΔCt (Spearman's ρ = 0.24, p = 0.131) was significantly associated with the classifier calibration score. Consistent with this, the single unclassified FFPE case had a DIN (4.2) and ΔCt (0.4) within the cohort's normal range, indicating that its failure was not attributable to DNA quality (Supporting Information S2: Table 3).
3.1. Methylation Classifier Performance
Methylation classifier performance was evaluated after stratifying cases according to tissue type and sampling approach.
3.1.1. Group 1: FFPE Cohort (n = 42)
Methylation classifier outputs were concordant with the initial histopathological diagnosis in 7% (3/42) of cases and refined the diagnosis in 48% (20/42). In 43% (18/42) of cases, the classifier output suggested an alternative classification relative to the original histological interpretation. Only 2% (1/42) of cases were unsatisfactory due to low classifier scores or lack of a confident match (Figure 1A), indicating a high rate of diagnostically informative classification in FFPE tissue. Block age was not associated with classification outcome: all six blocks older than 4 years (2017–2021) were successfully classified, and the single unclassified FFPE case derived from a recent (2023) block (Supporting Information S1: Table 2).
Figure 1.
Pie charts depicting methylation‐classifier prediction status stratified by tissue type (A) FFPE (n = 42), (B) Fresh Frozen: Histology Validated Sampling (n = 87), (C) Fresh Frozen: Blind Sampling (n = 50).
3.1.2. Group 2: Fresh‐Frozen—Histology‐Validated Sampling (n = 87)
When adequate tissue allowed prior histological verification, classifier performance was robust. Concordance with histopathology was observed in 52% (45/87) of cases, 33% (29/87) showed diagnostic refinement, and 15% (13/87) were reassigned to a new methylation‐based diagnosis relative to histology (Figure 1B). Notably, no cases were unclassified, highlighting the reliability of classification when tissue adequacy permitted histological validation.
3.1.3. Group 3: Fresh‐Frozen—Blind Sampling (n = 50)
Specimens with limited tissue availability (small/medium), which precluded prior histological validation, showed reduced classifier performance. Concordant classifications were seen in 28% (14/50) of cases, 38% (19/50) showed diagnostic refinement, and 12% (6/50) received a new methylation‐based diagnosis. However, 22% (11/50) of cases remained unclassified (Figure 1C).
Importantly, all discordant classifications in the fresh‐frozen cohort originated from the blind sampling group, where tumors initially diagnosed as high‐grade neoplasms frequently clustered with non‐neoplastic or low‐grade methylation classes (e.g., control cortex, reactive microenvironment, pilocytic astrocytoma, or ganglioglioma) (Table 1). This likely reflects low tumor cellularity or admixture with normal brain tissue rather than technical assay failure. Methylation classification results were validated using both the DKFZ Brain Tumor Classifier and the NIH Methylscape platform, confirming concordant outputs across platforms. Overall, tissue adequacy and prior histological validation significantly influence methylation classifier performance, with the most reliable results observed in FFPE samples and histology‐validated frozen tissues.
Table 1.
Discordant cases in methylation‐based molecular classification of CNS tissue samples.
| S. no. | FFPE/frozen | Histopathological diagnosis report | Methylation Profiling Classification (DKFZ Brain Tumor Classifier) | Methylation Profiling Classification (NIH Methylscape Classifier) |
|---|---|---|---|---|
| 1 | Frozen | Medulloblastoma, WNT‐ activated NOS | MC inflammatory microenvironment (0.66) | MC control tissue, reactive tumor microenvironment (0.66) |
| 2 | Frozen | Dysembryoplastic neuroepithelial tumor | Not classified (no score ≥ 0.3) | MC control tissue, reactive tumor microenvironment (0.44) |
| 3 | Frozen | Epithelioid glioblastoma, IDH wildtype, CNS WHO grade 4 | MC control tissue, cerebral hemisphere (0.94) | Ganglioglioma (0.98) |
| 4 | Frozen | Dysembryoplastic neuroepithelial tumor, CNS WHO grade 1 | MC control tissue, reactive tumor microenvironment (0.96) | MC pilocytic astrocytoma, CORT hemispheric (0.98) |
| 5 | Frozen | Atypical teratoid/rhabdoid tumor, CNS WHO grade 4 | MC glioblastoma, IDH‐wildtype, mesenchymal subtype (0.30) | Ganglioglioma (0.33) |
| 6 | Frozen | Diffuse Pediatric‐type high‐grade glioma, H3 wildtype and IDH‐wildtype | Not classified (no score ≥ 0.3) | Intracranial mesenchymal tumors (ICMT) with FET‐CREB, subtype B (0.46) |
| 7 | Frozen | Diffuse high‐grade glioma, WHO Grade 4, NOS | MC pilocytic astrocytoma, midline (0.45) | Control tissue, reactive tumor microenvironment (0.99) |
| 8 | Frozen | Astrocytoma IDH mutant CNS WHO Grade 4 | MC control tissue, cerebral hemisphere (0.99) | Ganglioglioma (0.83) |
| 9 | Frozen | Diffuse high‐grade glioma, WHO Grade 4, NOS | MC control tissue, reactive tumor microenvironment (0.65) | Chordoma (0.21) |
| 10 | Frozen | BCOR‐rearranged sarcoma | Not classified (no score ≥ 0.3) | Intracranial mesenchymal tumors (ICMT) with FET‐CREB, subtype B (0.29) |
| 11 | Frozen | Chordoid Meningioma | MC control tissue, hemispheric cortex (0.92) | Control tissue, reactive tumor microenvironment (0.77) |
| 12 | FFPE | Diffuse high‐grade glioma, Grade 4, NOS | Not classified (no score ≥ 0.3) | Control inflammatory tissue (0.79) |
4. Discussion
DNA methylation‐based classification has become an important component of integrated molecular diagnosis in CNS tumors. Since the introduction of the DKFZ Brain Tumor Classifier, several studies have demonstrated its high diagnostic accuracy and clinical utility across diverse CNS tumor entities [1, 2, 3]. However, practical implementation in routine neuropathology laboratories raises important questions regarding the optimal tissue substrate and the influence of pre‐analytical variables on classifier performance.
In the present study, we evaluated methylation classifier performance across FFPE and fresh‐frozen specimens while specifically examining the impact of sampling strategy. Our findings demonstrate that classifier performance is influenced more by sampling strategy than by the tissue preservation method itself, and are most consistent with variable tumor representation within the analyzed specimen. FFPE specimens in our cohort demonstrated excellent classification success, with diagnostically informative results obtained in 98% of cases. This observation is consistent with multiple real‐world implementation studies in which FFPE material has shown robust performance despite concerns regarding DNA degradation following formalin fixation [11, 22]. In routine pathology workflows, FFPE blocks are typically selected from histologically verified tumor‐rich regions, allowing direct assessment of tumor cellularity prior to molecular testing. This pre‐analytical histological verification likely explains the consistently high classification success observed in FFPE samples.
A key observation of this study is that fresh‐frozen tissue performs equally well when tumor adequacy is confirmed histologically. In the histology‐validated frozen cohort (Group 2), methylation classification was successful in all cases, demonstrating that frozen tissue is not intrinsically inferior to FFPE material for methylation‐based diagnostics. Rather, reliable classifier performance depends on adequate tumor cellularity within the analyzed specimen.
In contrast, frozen samples processed without prior histological validation (Group 3) demonstrated a substantially higher rate of discordant or unclassified results. Importantly, all classification failures in the frozen cohort occurred within this blind‐sampling subgroup. Tumors initially diagnosed histologically as high‐grade neoplasms frequently clustered with non‐neoplastic or low‐grade methylation classes such as control cortex, reactive microenvironment, pilocytic astrocytoma, or ganglioglioma. These findings are consistent with sampling bias and dilution of tumor DNA by non‐neoplastic tissue, rather than limitations of the methylation classifier itself.
An important observation in our study is that discordant classifications can occur despite high classifier confidence scores. Five discordant cases demonstrated calibrated scores exceeding 0.9 on both the DKFZ Brain Tumor and NIH Methylscape Classifiers. Similar observations have been reported in previous validation studies, highlighting that high classifier confidence does not necessarily guarantee biological accuracy when tumor cellularity is insufficient [3, 22, 23]. These findings highlight the continued importance of correlating methylation classifier outputs with histopathological findings and tumor cellularity rather than as standalone indicators of diagnostic accuracy. Similar to our findings, Pickles and colleagues reported that a small proportion of high‐confidence classifier predictions required cautious interpretation due to inadequate tumor representation within the analyzed sample [22]. Similarly, population‐based analyses of diffuse gliomas using FFPE material have demonstrated very low rates of unclassified cases when tumor cellularity is adequate [16] (Table 2). Importantly, tumor‐type distribution across sampling groups was comparable and did not demonstrate systematic bias toward specific entities, supporting that the observed differences in classification performance are primarily related to sampling strategy and tumor representation rather than cohort composition.
Table 2.
Reported classification performance of DNA methylation profiling in CNS tumors using FFPE and fresh‐frozen tissue.
| Study | Year | Tumor type(s) | Total sample (n) | FF (n) | FFPE (n) | Platform | Classification performance | Discordant/unresolved/misleading |
|---|---|---|---|---|---|---|---|---|
| Priesterbach‐Ackley and colleagues | 2020 | Mixed CNS tumors | 502 | 17 | 485 | Illumina 850K/EPIC | Classified: 90% | Discordant: 2% |
| Unresolved: 8% | ||||||||
| Capper and colleagues | 2018 | Mixed CNS tumors | 1104 | n/s | n/s | Illumina 450K/EPIC | Classified: 91% | Unresolved: 8% |
| Misleading: < 1% | ||||||||
| Karimi and colleagues | 2019 | CNS tumors | 55 | 0 | 55 | Illumina 450K/EPIC | Classified: 84% | Unresolved: 16% |
| Jaunmuktane and colleagues | 2019 | Adult CNS tumors | 325 | 0 | 325 | Illumina 450K/EPIC | Classified: 98% | Unresolved: 2% |
| Ferreyra Vega and colleagues | 2021 | Adult diffuse LGG | 166 | 0 | 166 | Illumina EPIC | Classified: 94% | Unresolved: 6% |
| Alharbi and colleagues [24] | 2020 | Medulloblastoma | 41 | 0 | 41 | Illumina EPIC 450K | Classified: 97.5% | Unresolved: 2.5% |
| Schenkel and colleagues [25] | 2022 | Glioma | 48 | 0 | 48 | Illumina EPIC | Classified: 31% | Unresolved: 69% |
| Wenger and colleagues [26] | 2022 | Pediatric CNS tumors | 121 | 49 | 72 | Illumina EPIC | Classified: 92% | Unresolved: 8% |
| Feinberg‐Gorenshtein and colleagues [27] | 2025 | Mixed CNS tumors | 30 | 0 | 30 | Nanopore (ONT) | Classified: 90% | Unresolved: 10% |
| Pickles and colleagues | 2020 | Pediatric CNS tumors | 306 | 0 | 306 | Illumina EPIC | Classified: 80% | Misleading: 20% |
| Uhl and colleagues | 2025 | Neuro‐oncological cases | 23 | 0 | 23 | Illumina EPIC (version v11b4) | Classified: 74% | Unresolved: 26% |
| Our findings | 2025 | Mixed CNS tumor | 179 | 137 | 42 | Illumina EPIC/EPIC v.2 | Classified: | Unclassified: |
| Group 1: 98% | Group 1: 2% | |||||||
| Group 2: 100% | Group 2: 0% | |||||||
| Group 3: 78% | Group 3: 22% |
Abbreviation: n/s, not specified.
These findings underscore the critical importance of tumor cellularity and pre‐analytical histological verification in methylation‐based diagnostics. In our cohort, both FFPE tissue and histology‐validated fresh‐frozen specimens demonstrated highly reliable classifier performance, whereas frozen samples processed without prior histological assessment showed increased rates of discordant or unclassified results, most likely reflecting sampling bias and inadequate tumor representation. Taken together, our results indicate that sampling methodology rather than tissue preservation format is an important factor associated with successful methylation classification, and reinforce the importance of integrating classifier outputs with careful histopathological evaluation in routine CNS tumor diagnostics.
A limitation of this study relates to the heterogeneity of molecular characterization across the cohort. Although all cases were evaluated in accordance with CNS WHO 5th edition recommendations using a combination of histomorphology, immunohistochemistry, FISH, and Sanger sequencing, the extent of molecular workup was not uniformly comprehensive for every case, reflecting real‐world diagnostic practice over a prolonged study period rather than a uniformly contemporary integrated diagnostic framework. In addition, although DNA methylation analysis is a robust and widely accepted tool for CNS tumor classification, the frozen aliquots used for methylation profiling were obtained in a blinded manner, and tumor cellularity was not directly quantified in the specific fragments used for DNA extraction; although retrospective review confirmed adequate overall tumor cellularity (> 50%) in the corresponding specimens, intratumoral heterogeneity was inferred rather than measured. We therefore explicitly acknowledge that the precise cause of classifier failure in the blind‐sampling subgroup cannot be definitively determined from the current data. Reduced or variable tumor representation remains the most plausible explanation—supported by the confinement of all frozen‐cohort failures to the blind‐sampling group and by the clustering of these cases with non‐neoplastic or low‐grade methylation classes, but other pre‐analytical or biological factors cannot be formally excluded, and establishing causation would require prospective studies incorporating direct, aliquot‐level assessment of tumor fraction. Accordingly, methylation classifier outputs in this study should be interpreted as complementary to the integrated histopathological diagnosis rather than as a standalone diagnostic ground truth.
5. Conclusion
DNA methylation‐based classification represents a reliable and practical tool for integrated CNS tumor diagnostics. Our findings indicate that histological verification of tumor adequacy prior to sampling, rather than preservation format alone, is an important factor associated with classifier reliability, although the precise cause of classifier failure in blind‐sampled specimens could not be definitively determined from the present data. When these pre‐analytical variables are controlled, both FFPE and fresh‐frozen tissue provide robust and clinically informative diagnostic information. These results support the broader clinical implementation of methylation profiling while underscoring the continued importance of expert histopathological correlation and integrated diagnostic interpretation.
Author Contributions
Jyotsna Singh and Vaishali Suri conceptualized the study. Data were curated by Jyotsna Singh, Shabnam Mansoori, Swati Singh, Supriya Bhardwaj, Srinidhi Vasant, and Kirti Srivastava. Formal analysis was performed by Jyotsna Singh, Shabnam Mansoori, Bharathi Prabakaran N. S., Swati Singh, Supriya Bhardwaj, Kirti Srivastava, Ashwinee Kumar, Srinidhi Vasant, and Vaishali Suri, while the investigation was carried out by Jyotsna Singh, Shabnam Mansoori, Bharathi Prabakaran N. S., Swati Singh, Supriya Bhardwaj, Srinidhi Vasant, Kirti Srivastava, Swati Singh, Ashwinee Kumar, Swati Mahajan, and Vaishali Suri. Methodology was developed with contributions from Jyotsna Singh, Shabnam Mansoori, Bharathi Prabakaran N. S., Swati Singh, Supriya Bhardwaj, Srinidhi Vasant, Kirti Srivastava, and Vaishali Suri. Software and data processing were managed by Jyotsna Singh, Shabnam Mansoori, Swati Singh, Supriya Bhardwaj, and Srinidhi Vasant. Visualization and figure preparation were performed by Jyotsna Singh, Shabnam Mansoori, Swati Singh, Supriya Bhardwaj, Srinidhi Vasant, Ashwinee Kumar, Swati Mahajan, and Vaishali Suri. Jyotsna Singh and Vaishali Suri prepared the original draft, and all authors contributed to reviewing and editing the manuscript and Vaishali Suri provided essential resources, supervision, and technical guidance. All authors read and approved the final version of the manuscript.
Ethics Statement
This study was approved by the Institutional Ethics Committee, All India Institute of Medical Sciences (AIIMS), New Delhi (Ref. No: AIIMS/A00916/05.03.2024).
Consent
Informed consent was obtained from all individual participants (or their parents or legal guardians) included in the study.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Supporting File 1
Supporting File 2
Acknowledgments
The authors sincerely thank their technical staff for their meticulous efforts in performing the immunohistochemistry staining and molecular analyses. V.S. acknowledges and thanks the Indian Council of Medical Research (ICMR) for funding this study. During the preparation of this work, the author(s) used ChatGPT solely to correct grammatical errors. The graphical abstract image was generated using AI‐assisted tools. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication. The graphical abstract image was generated using AI‐assisted tools. This work was supported by research grants from the Indian Council of Medical Research (ICMR), India [Grant No. EM/Dev/CAR/01/0090/2024] to Dr. Vaishali Suri (Principal Investigator).
Data Availability Statement
The raw data supporting this study can be made available upon reasonable request by contacting the corresponding author.
References
- 1. Louis D. N., Perry A., Wesseling P., et al., “The 2021 WHO Classification of Tumors of the Central Nervous System: A Summary,” Neuro‐Oncology 23, no. 8 (2021): 1231–1251, 10.1093/neuonc/noab106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Louis D. N., Perry A., Reifenberger G., et al., “The 2016 World Health Organization Classification of Tumors of the Central Nervous System: A Summary,” Acta Neuropathologica 131, no. 6 (2016): 803–820, 10.1007/s00401-016-1545-1. [DOI] [PubMed] [Google Scholar]
- 3. Capper D., Jones D. T. W., Sill M., et al., “DNA Methylation–Based Classification of Central Nervous System Tumours,” Nature 555, no. 7697 (2018): 469–474, 10.1038/nature26000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Jaunmuktane Z., Capper D., Jones D. T. W., et al., “Methylation Array Profiling of Adult Brain Tumours: Diagnostic Outcomes in a Large, Single Centre,” Acta Neuropathologica Communications 7, no. 1 (2019): 24, 10.1186/s40478-019-0668-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. White C. L., Kinross K. M., Moore M. K., et al., “Implementation of DNA Methylation Array Profiling in Pediatric Central Nervous System Tumors,” Journal of Molecular Diagnostics 25, no. 10 (2023): 709–728, 10.1016/j.jmoldx.2023.07.003. [DOI] [PubMed] [Google Scholar]
- 6. Priesterbach‐Ackley L. P., Boldt H. B., Petersen J. K., et al., “Brain Tumour Diagnostics Using a DNA Methylation–Based Classifier,” Neuropathology and Applied Neurobiology 46, no. 5 (2020): 478–492, 10.1111/nan.12610. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Wu Z., Abdullaev Z., Pratt D., et al., “Impact of the Methylation Classifier and Ancillary Methods on CNS Tumor Diagnostics,” Neuro‐Oncology 24, no. 4 (2022): 571–581, 10.1093/neuonc/noab227. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Ahmad Z., Rahim S., Abdul‐Ghafar J., et al., “Events in CNS Tumor Pathology Post‐2016 WHO CNS,” International Journal of General Medicine 16 (2023): 107–127, 10.2147/IJGM.S394872. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Singh J., Sahu S., Mohan T., et al., “Current Status of DNA Methylation Profiling in Neuro‐Oncology,” Neuro‐Oncology Practice 10, no. 6 (2023): 518–526, 10.1093/nop/npad040. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Koelsche C. and von Deimling A., “Methylation Classifiers: Brain Tumors, Sarcomas, and What's Next,” Genes, Chromosomes and Cancer 61, no. 6 (2022): 346–355, 10.1002/gcc.23041. [DOI] [PubMed] [Google Scholar]
- 11. Sahm F., Schrimpf D., Stichel D., et al., “DNA Methylation‐Based Classification and Grading System for Meningioma: A Multicentre, Retrospective Analysis,” Lancet Oncology 18, no. 5 (2017): 682–694, 10.1016/S1470-2045(17)30155-9. [DOI] [PubMed] [Google Scholar]
- 12. Sill M., Schrimpf D., Patel A., et al., “Advancing CNS Tumor Diagnostics With Expanded DNA Methylation‐Based Classification,” preprint, medRxiv, May 29, 2025, 10.1101/2025.05.28.25328344. [DOI] [PubMed]
- 13. Aldape K., Capper D., von Deimling A., et al., “cIMPACT‐NOW Update 9: Recommendations on Utilization of Genome‐Wide DNA Methylation Profiling for CNS Tumor Diagnostics,” Neuro‐Oncology Advances 7, no. 1 (2025): vdae228, 10.1093/noajnl/vdae228. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Kling T., Wenger A., Beck S., and Carén H., “Validation of the MethylationEPIC BeadChip for Fresh‐Frozen and FFPE Tumours,” Clinical Epigenetics 9 (2017): 33, 10.1186/s13148-017-0333-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Klughammer J., Kiesel B., Roetzer T., et al., “The DNA Methylation Landscape of Glioblastoma Disease Progression,” Nature Medicine 24, no. 10 (2018): 1611–1624, 10.1038/s41591-018-0156-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Ferreyra Vega S., Olsson Bontell T., Corell A., Smits A., Jakola A. S., and Carén H., “DNA Methylation Profiling for Molecular Classification of Adult Diffuse Lower‐Grade Gliomas,” Clinical Epigenetics 13, no. 1 (2021): 102, 10.1186/s13148-021-01085-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Pagès M., Uro‐Coste E., Colin C., et al., “Implementation of DNA Methylation Profiling Into a Multistep Diagnostic Process in Pediatric Neuropathology,” Cancers 13, no. 6 (2021): 1377, 10.3390/cancers13061377. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Schepke E., Löfgren M., Pietsch T., et al., “DNA Methylation Profiling Improves Routine Diagnosis of Paediatric CNS Tumours,” Neuropathology and Applied Neurobiology 48, no. 6 (2022): e12838, 10.1111/nan.12838. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Karimi S., Zuccato J. A., Mamatjan Y., et al., “The CNS Tumor Methylation Classifier Changes Neuro‐Oncology Practice for Challenging Diagnoses and Impacts Patient Care,” Clinical Epigenetics 11, no. 1 (2019): 185, 10.1186/s13148-019-0766-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Uhl C., Ayoub N., Faust K., et al., “DNA Methylation–Based Profiling in Unspecified Neuro‐Oncological Cases,” Brain Spine 5 (2025): 104256, 10.1016/j.bas.2025.104256. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Koelsche C. and von Deimling A., “Methylation Classifiers: Brain Tumors, Sarcomas, and What's Next,” Genes, Chromosomes and Cancer 61, no. 6 (June 2022): 346–355. [DOI] [PubMed] [Google Scholar]
- 22. Pickles J. C., Fairchild A. R., Stone T. J., et al., “DNA Methylation–Based Profiling for Paediatric CNS Tumour Diagnosis,” Lancet Child & Adolescent Health 4, no. 2 (2020): 121–130, 10.1016/S2352-4642(19)30342-6. [DOI] [PubMed] [Google Scholar]
- 23. Morgacheva D., Ryzhova M., Zheludkova O., Belogurova M., and Dinikina Y., “DNA Methylation‐Based Diagnosis Confirmation in a Pediatric Patient With Low‐Grade Glioma: A Case Report,” Frontiers in Pediatrics 11 (2023): 1256876, 10.3389/fped.2023.1256876. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Alharbi M., Mobark N., Bashawri Y., et al., “Methylation Profiling of Medulloblastoma in a Clinical Setting Permits Sub‐Classification and Reveals New Outcome Predictions,” Frontiers in Neurology 11 (2020): 167, 10.3389/fneur.2020.00167. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Schenkel L. C., Mathew J., Hirte H., et al., “Evaluation of DNA Methylation Array for Glioma Tumor Profiling,” Genes 13, no. 11 (2022): 2075, 10.3390/genes13112075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Wenger A., Ferreyra Vega S., Schepke E., et al., “DNA Methylation Alterations Across Time and Space in Paediatric Brain Tumours,” Acta Neuropathologica Communications 10, no. 1 (2022): 105, 10.1186/s40478-022-01406-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Feinberg‐Gorenshtein G., Grunwald A., Vermeulen C., et al., “Brain Tumor Classification From FFPE Samples Using Nanopore Methylation Sequencing,” NAR Cancer 7, no. 4 (2025): zcaf038, 10.1093/narcan/zcaf038. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting File 1
Supporting File 2
Data Availability Statement
The raw data supporting this study can be made available upon reasonable request by contacting the corresponding author.