One Nanopore Panel Reads Out Parkinson’s Genes and Repeat Expansions Together

Genetic testing for Parkinson’s disease has long been a piecemeal affair. Clinicians order one assay for single-nucleotide variants, another for large structural rearrangements, and yet another for the repeat expansions that underlie ataxias, Huntington’s disease, and amyotrophic lateral sclerosis. Each test adds cost, delay, and uncertainty. Now a team led by researchers at the Institute of Neurogenetics at the University of Lübeck, together with collaborators in Austria, Turkey, the Philippines, Canada, and Germany, reports that a single targeted long-read sequencing panel can interrogate all of these variant classes at once, with an overall detection rate of 85 percent for known pathogenic variants. The study, published in npj Parkinson’s Disease, offers a detailed technical proof of concept for a diagnostic approach that could eventually collapse a fragmented testing pipeline into one run.

The panel itself is ambitious in scope. The researchers assembled 564 gene loci linked to Parkinson’s disease, repeat expansion disorders, and other neurodegenerative movement disorders, drawing on established diagnostic panels from the genomics company CeGaT, the International Parkinson and Movement Disorder Society Taskforce, 90 Parkinson’s disease genome-wide association study loci, and the catalog of known repeat expansion genes. Each gene locus was padded with 50 kilobases of flanking sequence on either side, and the mitochondrial genome was added as a separate target. Together the target regions span roughly 113.6 million base pairs, or about 3.44 percent of the GRCh38 human reference genome, comfortably within the 1 to 5 percent fraction recommended for optimal enrichment with adaptive sampling.

The sequencing strategy relies on Oxford Nanopore Technologies’ adaptive sampling, a clever real-time filtering technique built into the MinKNOW operating software. As DNA molecules thread through the protein nanopores, the software maps the first few hundred bases of each read against a predefined BED file of target coordinates. Reads that align to the panel’s genes continue to be sequenced in full, while off-target molecules are ejected from the pore, freeing it to capture another molecule. The result is targeted enrichment without any wet-lab capture step, preserving the native DNA and its long read lengths. In this study, three patient samples were multiplexed per PromethION flow cell and sequenced for 72 hours using R10.4.1 flow cells and super-accuracy base calling with the Dorado basecaller.

The performance metrics are respectable for a first demonstration. Across 18 individuals, the panel achieved a mean on-target coverage of 24X per sample, with a standard deviation of 9X, and a mean on-target read N50 of 21.5 kilobases, a read-length metric considered suitable for detecting structural variants. The average Phred quality score was 19.7. Coverage varied considerably across loci, partly because some genes sit in notoriously difficult genomic terrain: several panel genes lie within or near the highly polymorphic major histocompatibility complex on chromosome 6, and 20 of the 31 genes with mean coverage below 15X are X-linked and therefore less abundant in the ten male participants. The mitochondrial genome, by contrast, reached a mean depth of 1256X.

To test diagnostic utility, the team sequenced 15 positive controls with previously characterized pathogenic variants and 3 negative controls with idiopathic Parkinson’s disease or frontotemporal dementia. The positive controls spanned the full variant spectrum: single-nucleotide variants and small indels in GBA1, LRRK2, PRKN, and RAB32; large deletions in PRKN ranging from 20 to 320 kilobases; a 1.7-megabase triplication encompassing the entire SNCA gene; and repeat expansions in ATXN1, ATXN2, ATXN3, HTT, DAB1, FGF14, C9orf72, and the TAF1-SVA retrotransposon insertion that causes X-linked dystonia-parkinsonism. Of the 20 expected pathogenic variants, 17 were clearly identified, a detection rate of 85 percent with a Clopper-Pearson 95 percent confidence interval of 62 to 97 percent.

The small-variant results were flawless. The EPI2ME human variation workflow, which chains together Minimap2 alignment, the Clair3 deep-learning variant caller, the Sniffles2 structural variant caller, and the Straglr repeat caller, correctly detected all six known single-nucleotide variants and small indels, including the LRRK2 p.G2019S mutation and three PRKN variants. This success reflects years of improvement in nanopore base-calling accuracy, with Q-scores above 20 now routine using R10 chemistry. Structural variants proved harder. Sniffles2 with default parameters caught two large PRKN deletions but missed three others, and no caller detected the SNCA triplication. However, the alternative callers CuteSV and SVIM, run with adjusted size parameters, recovered two of the missed deletions, and visual inspection in the Integrative Genomics Viewer confirmed all PRKN deletions, enabling the team to establish compound heterozygosity through read-based phasing with WhatsHap. The SNCA triplication left a telltale signature even without a formal call: coverage over the SNCA locus was 2.5-fold higher than the panel average, a hint that dedicated copy-number tools for targeted data could convert into a definitive diagnosis.

Repeat expansions showed a mixed picture. Straglr correctly identified pathogenic expansions in ATXN1, ATXN2, ATXN3, and HTT, with repeat numbers closely matching prior results, and the independent Noise Cancelling Repeat Finder tool confirmed them with deviations of only a few repeat units. Expansions in DAB1 and FGF14 were missed by the workflow version of Straglr simply because those genes are absent from its reference catalog, but a standalone version with a custom input file, together with NCRF, detected both. The two failures are instructive. The C9orf72 hexanucleotide expansion, whose GC-rich repeats can form G-quadruplex structures, yielded only six expanded reads despite 43X coverage, too few and too variable to assign a reliable repeat number. The TAF1-SVA insertion is absent from the GRCh38 reference genome entirely, so adaptive sampling discarded some relevant reads as off-target, and the gene’s X-chromosome location further reduced coverage in male samples, leaving only three informative reads. Both cases were classified as inconclusive rather than false negatives, and the authors suggest that reducing the target fraction or increasing coverage could resolve them.

Beyond confirming known variants, the panel surfaced five additional findings in five patients that had not been previously reported for those individuals: potentially pathogenic variants in STXBP2, AP4S1, RARS2, and ALS2, and a CCTG repeat expansion of 137 units in CNBP, the gene associated with myotonic dystrophy type 2. All were validated by Sanger sequencing or PCR and gel electrophoresis. Notably, none of the affected patients showed clinical features of the corresponding conditions, which are mostly recessive, so the team interprets these as incidental findings demanding cautious counseling rather than new diagnoses. Intriguingly, 13 patients also carried benign AAAAG expansions in RFC1, below the pathogenic threshold of 250 repeats, illustrating how comprehensively long reads expose the repetitive landscape of the genome.

The economics are part of the appeal. With three samples multiplexed per flow cell, the sequencing cost per patient, roughly 330 to 830 euros, matches a whole-genome nanopore run for a single individual, and validation steps would bring the total to roughly 500 to 1000 euros as a first-line test. Because the panel is defined by a simple text file of genomic coordinates, it can be updated overnight as new Parkinson’s risk loci and repeat motifs are discovered. The authors envision it as either a first-line combined test ordered by an experienced clinician or a second-line assay after negative short-read sequencing. Limitations remain real: the cohort was small, four samples came from induced pluripotent stem cells rather than blood, which is unsuitable for clinical diagnostics because reprogramming can introduce somatic variants, and most variant callers are still optimized for whole-genome rather than targeted long-read data. Manual inspection of difficult variants, about 30 minutes per sample, is not scalable. Still, as specialized tools mature and larger validation cohorts accumulate, the vision of a single, flexible, cost-effective test that reads out point mutations, massive rearrangements, and sprawling repeat expansions in one pass is coming sharply into focus.

Subject of Research: A targeted long-read nanopore sequencing gene panel for the combined genetic diagnosis of Parkinson's disease and repeat expansion disorders

Article Title: Unified long-read panel for Parkinson’s and repeat expansion disorders

Article References: Fienemann, A., Prietzsche, J. C., Laβ, J., Much, C., Schaake, S., Lüth, T., Gabbert, C., Zimprich, A., Stögmann, E., König, T., Ganos, C., Gül-Demirkale, T., Başak, A. N., Jamora, R. D. G., Rosales, R. L., Saranza, G., Diesta, C. C. E., Möller, M., Borsche, M., … Trinh, J. (2026). Unified long-read panel for Parkinson’s and repeat expansion disorders. npj Parkinson's Disease, 12(1), Article 233. https://doi.org/10.1038/s41531-026-01585-4

Image Credits: AI Generated

DOI: 10.1038/s41531-026-01585-4

Keywords: Parkinson's disease, long-read sequencing, Oxford Nanopore, adaptive sampling, repeat expansion disorders, genetic diagnostics, structural variants, LRRK2, PRKN, SNCA, C9orf72, gene panel