Nanopore Sequencing Delivers Full Pediatric Brain Tumor Workup in a Single Assay

Diagnosing a childhood brain tumor currently means running several separate molecular tests: DNA methylation arrays, short-read sequencing of key genes, and sometimes RNA sequencing, each with its own cost, delay, and sample requirement. A new study published in Acta Neuropathologica Communications suggests that a single long-read nanopore sequencing run could replace much of that fragmented workflow, delivering methylation-based classification, copy-number profiles, and variant detection from one assay, while also revealing where the technology still falls short of established methods.

Researchers led by Sandra Ferreyra Vega and Helena Carén at the University of Gothenburg evaluated 20 tumor samples from 20 children treated at Sahlgrenska University Hospital in the Västra Götaland region of Sweden, diagnosed between 2013 and 2025. The cohort comprised eight gliomas, five ependymomas, and seven medulloblastomas, all classified according to the 2021 WHO CNS5 criteria. Every sample had matched Illumina short-read sequencing and DNA methylation array data from routine clinical work-up, allowing direct, sample-by-sample comparison with the nanopore results.

The team sequenced each tumor individually on a PromethION 2 Solo instrument for 72 hours. Each run yielded on average 126 gigabases of sequence, corresponding to a mean genome coverage of approximately 40-fold, with reads averaging an N50 length of 7.79 kilobases. Coverage varied between tumor types, with the greatest variability among diffuse gliomas, followed by ependymomas and medulloblastomas, while pilocytic astrocytomas showed relatively uniform coverage across samples.

Copy-number profiling was the strongest performance area. Nanopore-derived CNV profiles showed a median Pearson correlation of 0.816 with methylation array-derived profiles, with an interquartile range of 0.683 to 0.935. To confirm that this agreement was not an artifact, the researchers compared each array profile against all 20 nanopore profiles: the matched profile correlated far more strongly than mismatched profiles (median r = 0.816 versus 0.111), and it ranked highest in 18 of 20 cases and within the top three in 19 of 20. The analysis also showed that copy-number-flat profiles, which offer little dynamic range, produced lower correlations even when the actual signal differences were tiny, as in one case with a correlation of 0.36 but a mean absolute signal difference of just 0.034.

Beyond conventional copy-number analysis, the long reads enabled haplotype-specific profiling through phasing, distinguishing copy-number changes affecting individual parental chromosomes and revealing allelic imbalance and loss of heterozygosity that are not directly apparent from standard profiles. This capability, the authors note, provides information beyond what conventional array-based CNV analysis can offer.

Variant detection showed a more nuanced picture. Of 78 clinically relevant small variants detected by Illumina sequencing, 74 (95 percent) had some form of nanopore evidence, but only 53 (68 percent) were recovered as exact high-confidence calls. Recovery was markedly better for single-nucleotide variants (51 of 68, or 75 percent) than for insertions and deletions (2 of 10, or 20 percent), although all ten indels showed nanopore evidence after additional computational or manual review. Only four Illumina consensus target variants, or 5 percent, remained without detectable nanopore evidence, and three of those had low variant allele frequencies of 5 to 12 percent. Conversely, of 22 high-confidence nanopore variants absent from the Illumina consensus, 17 had some Illumina support upon closer inspection, and only five lacked any detectable Illumina evidence. Recovery of variants rose with increasing variant allele frequency, and the NF1 gene stood out as a particular challenge, where discordant variants were mostly supported by reads with near-zero mapping quality, pointing to long-read alignment difficulties rather than missing signal.

Structural variant detection was the weakest area of cross-platform agreement. Nanopore sequencing identified far more deletions, duplications, and insertions than Illumina (924, 683, and 786 versus 292, 249, and 9), while Illumina detected more breakend events (1,797 versus 1,085). The size distributions differed dramatically, with median Illumina deletion and duplication sizes of 15.0 and 50.6 kilobases compared with 956 base pairs and 2.0 kilobases for nanopore. Breakpoint overlap was low for most sample pairs: in 19 of 20 samples, complete two-breakpoint support ranged from 0 to 15.5 percent for Illumina-derived events and 0 to 11.1 percent for nanopore-derived events, with one highly rearranged tumor as an exception at 53.0 and 27.5 percent. Because the two technologies were applied to spatially distinct tumor samples, the study cannot separate technical differences from spatial intratumor heterogeneity.

Methylation-based classification was a highlight. Using six nanopore-based classifiers, the researchers found that methylation-class assignments agreed with the integrated WHO CNS5 diagnoses in 90 to 100 percent of samples, depending on the classifier: 100 percent for MNP-Flex and Rapid-CNS2, 95 percent for M-PACT, and 90 percent for NanoDx, PanNanoDx, and Sturgeon. Ependymomas, medulloblastomas, pilocytic astrocytomas, and one IDH-mutant astrocytoma were classified concordantly across all six classifiers, although not every prediction reached the classifier-specific confidence thresholds. Discordant classifications were confined to three diffuse high-grade glioma samples, including two associated with constitutional mismatch repair deficiency, a rare subtype that is likely underrepresented in classifier training datasets. Notably, classifier confidence scores did not directly reflect diagnostic agreement, as some below-threshold classifications were still concordant while occasional discordant calls came with high confidence scores.

The downsampling experiments offered perhaps the most clinically consequential finding. When the researchers generated reduced-depth datasets at approximately 1-, 5-, and 10-fold genome coverage, concordance with WHO CNS5 diagnoses remained largely stable, ranging from 90 to 100 percent across classifiers and depths. The proportion of classifications reaching high-confidence thresholds varied more substantially and was classifier dependent, with M-PACT maintaining high confidence at all depths while PanNanoDx showed the lowest rates that improved with coverage. Based on the observed sequencing yield, the estimated times to reach 1-, 5-, and 10-fold coverage were approximately 1.8, 9, and 18 hours, corresponding to projected methylation-classification results after roughly 2, 9 to 10, and 19 hours. Comprehensive analysis at approximately 40-fold coverage, including the full Lumos analysis pipeline of about 8.1 hours, was estimated to require approximately 3 to 4 days from sequencing initiation. The authors stress these are computational projections from retrospective data, not prospectively measured clinical turnaround times.

The study’s authors are careful to frame the work as a diagnostic evaluation rather than a validation for clinical deployment. The cohort is small and retrospective, covering only three major tumor groups, which limits generalizability to rarer pediatric entities. The nanopore variants were called with a tumor-only workflow while the Illumina reference used tumor-normal sequencing, and the reduced-depth analysis was performed computationally rather than in real time. The researchers conclude that prospective studies in larger, diagnostically diverse cohorts are needed to establish performance and turnaround time in routine clinical practice, but their findings support the feasibility of integrating multiple diagnostic modalities into a single nanopore-based workflow, with shorter sequencing runs potentially delivering methylation classifications hours rather than days after a tumor reaches the laboratory.

Subject of Research: Evaluation of long-read nanopore sequencing for molecular diagnosis of pediatric CNS tumors

Article Title: Diagnostic evaluation of long-read nanopore sequencing for pediatric CNS tumors

Article References: Ferreyra Vega, S., Kwan, R. S.-Y., Rahmqvist, I., Schepke, E., Sabel, M., Olsson Bontell, T., Kling, T., & Carén, H. (2026). Diagnostic evaluation of long-read nanopore sequencing for pediatric CNS tumors. Acta Neuropathologica Communications, 14(1), Article 224. https://doi.org/10.1186/s40478-026-02466-w

Image Credits: AI Generated

DOI: 10.1186/s40478-026-02466-w

Keywords: nanopore sequencing, pediatric CNS tumors, DNA methylation, copy-number variation, structural variants, medulloblastoma, glioma, ependymoma, WHO CNS5 classification, molecular diagnostics, PromethION, turnaround time