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

The logistics of a motor neuron are staggering. These cells extend axons that can stretch up to a meter in the human body, and their distant terminals need a constant supply of fresh mitochondria, the organelles that generate the chemical energy ATP required for synaptic transmission and cellular survival. Moving mitochondria along the axon is an energy-intensive process in itself, and when that transport falters, the far reaches of the neuron are starved of power while damaged organelles pile up. Disrupted axonal transport has been implicated in ALS for years, and previous work had established that C9ORF72 motor neurons show cell-autonomous defects in mitochondrial motility and function. What remained unknown was whether the astrocytes surrounding those neurons—cells already known to contribute to non-cell-autonomous neurodegeneration through mechanisms such as reduced glutamate uptake and impaired lactate shuttling—also influence this transport machinery.

To find out, the team built a rigorous human stem-cell platform. They used induced pluripotent stem cell lines derived from three patients carrying the C9ORF72 repeat expansion, together with isogenic gene-corrected controls created through CRISPR–Cas9 editing, in which the mutation was removed. From these lines they generated highly enriched spinal cord-patterned astrocytes, more than ninety percent of which expressed the canonical astrocyte markers glial fibrillary acidic protein and S100B, as well as motor neuron cultures of which roughly half to sixty percent were ISL1/2-positive motor neurons. The mutant astrocytes carried hallmark C9ORF72 pathologies, including RNA foci and dipeptide repeat proteins, yet showed no differences in glutamate uptake or calcium wave propagation, meaning any effects on transport could not be attributed to those well-known astrocyte dysfunctions.

The central experiment was elegantly simple. The researchers selectively labeled motor neurons with a fluorescent mitochondrial marker, mito-DsRed2, and then imaged mitochondrial movement along a hundred-micrometer stretch of the proximal axon using live-cell time-lapse microscopy, generating kymographs from which they extracted two established metrics: the percentage of mitochondria that were motile and their average velocity. When genetically normal motor neurons were grown in physical contact with C9ORF72-mutant astrocytes, both measures dropped significantly compared with motor neurons cultured alone. Strikingly, when the same control neurons were paired with gene-corrected astrocytes, the deficit vanished entirely. Even more remarkable, mutant motor neurons cocultured with corrected astrocytes recovered both transport parameters to wild-type levels. In these mixed cultures, the phenotype of the neuron’s mitochondrial traffic was dictated not by the neuron’s own genome but by the genotype of its astrocyte neighbors.

Contact, it turned out, was essential. When the team treated control motor neurons with conditioned medium harvested from mutant astrocytes, mitochondrial motility was unaffected, indicating that secreted factors alone could not reproduce the damage; direct physical interaction between the two cell types was required. The specificity of the defect was equally revealing. Tracking HcT-labeled signaling endosomes, another vital axonal cargo, showed no transport changes in any coculture combination. The astrocyte effect was therefore cargo-specific, targeting mitochondria rather than causing a global collapse of axonal transport. The authors suggest this selectivity reflects distinct energetic dependencies: signaling endosomes fuel their movement with locally generated glycolytic ATP, whereas mitochondrial motility depends on membrane potential and calcium-regulated adaptors such as Miro, directly coupling transport to the bioenergetic state of the organelle itself.

That coupling became the thread the researchers followed next. Using Seahorse extracellular flux analysis, they measured oxygen consumption rates as a proxy for mitochondrial respiration. Because motor neuron cultures respire at nearly eightfold the rate of astrocytes alone, changes in coculture readings could be confidently attributed to the neurons. The results were unambiguous: coculture with astrocytes of either genotype boosted basal and maximal respiration relative to neuron monocultures, but pairing neurons with mutant astrocytes significantly reduced both parameters compared with pairing them with corrected astrocytes. A complementary assay using the potentiometric dye MitoTracker Red CMXRos confirmed that the steady-state mitochondrial membrane potential of motor neurons—mutant or normal—fell in the presence of mutant astrocytes and recovered with corrected astrocytes. The astrocyte genotype was directly modulating the bioenergetic health of neurons it touched.

The root of the problem lay within the astrocytes themselves. Isolated C9ORF72-mutant astrocytes showed significantly reduced basal, ATP-linked and maximal respiration compared with their gene-corrected counterparts, along with diminished glycolytic capacity, lactate production and glucose utilization. RNA sequencing, however, revealed no genotype-driven astrocyte reactivity, making a gain-of-toxic-function inflammatory mechanism unlikely. Intriguingly, when the team measured intracellular lactate in motor neurons using a genetically encoded FRET biosensor called Laconic, they found no significant differences across coculture conditions, suggesting that disrupted oxidative metabolism within astrocytes, rather than altered lactate transfer to neurons, mediates the transport deficit. The astrocytes were not failing to deliver fuel; they were failing in a way that somehow drains the neurons’ own power plants.

The therapeutic implication emerged from a rescue experiment targeting the master regulator of mitochondrial biogenesis. The team overexpressed three key metabolic genes in mutant astrocytes via lentiviral delivery: PPARGC1A (encoding PGC-1α), PPARGC1B (PGC-1β) and TFAM. PGC-1β proved toxic to the astrocytes, and TFAM, consistent with prior reports that it does not drive mitochondrial biogenesis, conferred no benefit. PGC-1α, by contrast, upregulated transcripts of the mitochondrial electron transport chain, including MT-CO2, MT-ND2, MT-CO3, MT-ND4 and MT-ATP6, and produced a trend toward increased respiration. When mutant astrocytes overexpressing PGC-1α were cocultured with motor neurons—mutant or corrected—mitochondrial motility and velocity were restored. Boosting the metabolic engine of the astrocyte alone was sufficient to repair the transport machinery inside genetically distinct neurons.

The findings reframe how scientists think about non-cell-autonomous degeneration in ALS. Rather than astrocytes poisoning neurons through inflammatory or excitotoxic signals, this work describes a bioenergetically coupled mechanism in which the metabolic competence of glia sets the ceiling for mitochondrial dynamics in neurons. The result also resonates with earlier animal work in SOD1 models of ALS, where PGC-1α activation preserved mitochondrial function and delayed disease progression, hinting that metabolic rescue of glia could be a broadly applicable strategy. The authors caution that much remains to be defined, including whether direct mitochondrial exchange, metabolite or lipid transfer, vesicular signaling or other modes of metabolic coupling connect astrocytic energy status to neuronal mitochondrial behavior, and whether similar mechanisms operate across other ALS genotypes and in sporadic disease, which accounts for roughly ninety percent of cases.

For a disease with no cure and only modestly effective therapies, the identification of an actionable, astrocyte-centered node is significant. If pharmacological activation of the PGC-1α pathway in glia can correct non-cell-autonomous mitochondrial dysfunction in C9ORF72-ALS, it opens a therapeutic window that does not require correcting the mutation inside every neuron. The study, funded in part by the Medical Research Council, the UK Dementia Research Institute and the My Name’5 Doddie Foundation among others, demonstrates with unusual clarity that in neurodegeneration, the health of a neuron may depend as much on the metabolic company it keeps as on its own genes—a lesson that could reshape drug discovery for ALS well beyond the C9ORF72 mutation that first revealed it.

Subject of Research: Non-cell-autonomous regulation of axonal mitochondrial transport by C9ORF72-mutant astrocytes in amyotrophic lateral sclerosis

Article Title: Astrocytes regulate axonal mitochondrial transport deficits in C9ORF72 amyotrophic lateral sclerosis motor neurons

Article References: Stavrou, M., Heffernan, Á. B., Carter, R. N., Dando, O., Jiwaji, Z., Burr, K., Nanda, J., Villarroel-Campos, D., Masoud Abdelhafid, A., Cholewa-Waclaw, J., Soong, D., Story, D., Schiavo, G., Hardingham, G. E., Chandran, S., & Selvaraj, B. T. (2026). Astrocytes regulate axonal mitochondrial transport deficits in C9ORF72 amyotrophic lateral sclerosis motor neurons. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02464-0

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02464-0

Keywords: amyotrophic lateral sclerosis, C9ORF72, astrocytes, motor neurons, mitochondrial transport, axonal transport, mitochondrial bioenergetics, PGC-1α, induced pluripotent stem cells, non-cell-autonomous neurodegeneration, CRISPR gene correction, live-cell imaging

Phase separation is the physical phenomenon by which a well-mixed solution splits into distinct phases, the way oil separates from vinegar in a shaken bottle of salad dressing. In cells, proteins containing intrinsically disordered, low-complexity domains can drive the formation of membraneless compartments called condensates, which concentrate specific molecules without any enclosing lipid membrane. Nucleoli, stress granules and nuclear speckles all form this way. The key property of healthy condensates is dynamism: molecules diffuse in and out, and the entire assembly can dissolve within seconds or minutes once conditions change. This liquid-like behavior depends on weak, multivalent interactions among the low-complexity domains, which constantly break and reform.

The trouble begins when those weak interactions stop being weak and transient. The review describes how, in ALS, condensates formed by RNA-binding proteins such as FUS and TDP-43 progressively lose their reversibility, maturing from liquid droplets into gel-like states and eventually into solid fibrillar aggregates that are hallmarks of the diseased motor neuron. This liquid-to-solid transition, sometimes called pathological hardening, is driven by the same low-complexity domains that make condensates form in the first place. Given time, the transient beta-sheet-like contacts within these domains can lock into stable structures, converting a dynamic organelle into an irreversible inclusion that the cell cannot dismantle.

What makes the new review distinctive is its emphasis on the layers of cellular regulation that normally prevent this maturation. The authors assemble evidence that molecular chaperones, best known for helping newly made proteins fold, also patrol condensates. Heat shock protein 70 family members, for example, can bind exposed hydrophobic stretches within condensates, remodel their internal organization and keep the assemblies fluid. Chaperone activity thereby extends the window in which a condensate remains reversible, buying the cell time to dissolve it before aggregation becomes permanent. When chaperone capacity is overwhelmed or declines, as it does with age and stress, that window narrows.

Nuclear import receptors form a second, perhaps more surprising, line of defense. Karyopherin-beta2, the receptor that ferries FUS into the nucleus, does more than transport. It binds directly to the proline-tyrosine nuclear localization signal within FUS and, in doing so, physically shields the low-complexity domain from engaging in the multivalent contacts that drive phase separation and fibrillization. In effect, the import receptor acts as a portable anti-aggregation chaperone, keeping FUS soluble during its journeys through the cytoplasm. The review highlights how this mechanism illustrates a broader principle: compact folded domains can modulate the behavior of disordered regions they are physically linked to or bound against.

The proteostasis network supplies a third layer of regulation. Enzymatic components such as protein disulfide isomerase and peptidyl-prolyl cis-trans isomerase A can alter the conformational state of proteins inside or near condensates, while ubiquilin-like protein 2, an ALS-linked protein, helps shuttle ubiquitinated cargo between condensates and the protein degradation machinery. Optineurin, another ALS-associated factor, participates in autophagic clearance of aggregates. Together these systems form a quality-control pipeline that continuously remodels and removes material from condensates, limiting the residence time of any given molecule and thereby reducing the probability that transient contacts will ripen into stable aggregates.

The review then turns to one of the most devastating genetic insults in ALS: the hexanucleotide repeat expansion in the C9ORF72 gene, the most common known cause of familial disease. This expansion produces arginine-rich dipeptide repeat proteins, and these short, highly basic peptides are potent modifiers of phase behavior. They can seed the condensation of RNA-binding proteins and, critically, impair nuclear import pathways, including the transport receptors that keep FUS in check. The result is a double hit: more condensation is promoted at the same time as the systems that would normally reverse it are disabled. The authors present this as a mechanistic explanation for why C9ORF72 pathology so closely resembles the FUS and TDP-43 proteinopathies seen across the ALS spectrum.

Perhaps the most forward-looking section of the review concerns zinc finger domains. Recent structural and biophysical studies suggest that these small, compact, folded modules can recognize specific polymer states of low-complexity domains and constrain their growth. Rather than dissolving condensates outright, zinc finger domains appear to cap or crosslink the polymers in ways that limit further elongation and material transition. This supports what the authors describe as a model in which structured domains act as modulators of higher-order assemblies, reading the physical state of a condensate and restraining its maturation. If confirmed across multiple ALS-associated proteins, this recognition mechanism could represent an entirely new regulatory vocabulary that cells use to manage their liquid organelles.

The therapeutic implications are considerable. Most ALS drug development has targeted downstream events, such as oxidative stress, excitotoxicity or the aggregates themselves. The framework laid out by the Nara Medical University team suggests an upstream strategy: preserve or restore condensate homeostasis. Enhancing chaperone activity, boosting nuclear import, supporting proteostasis components or mimicking the polymer-capping behavior of zinc finger domains could all, in principle, keep pathological proteins in their reversible, harmless state before fibrils ever form. Because the same regulatory principles appear to govern TDP-43, FUS and C9ORF72-related pathology, a therapy aimed at condensate homeostasis might cut across the genetic subtypes of the disease rather than addressing them one by one.

Cautious optimism is warranted. The review is a synthesis of mechanistic studies rather than a demonstration of efficacy in patients, and the authors themselves frame their conclusions as a model to be tested: clarifying how chaperones, import receptors, proteostasis enzymes and structured domains operate across ALS-associated proteins remains the task ahead. Yet the conceptual shift is hard to overstate. ALS, on this view, is not merely a disease of misfolded proteins but a disease of lost reversibility, in which the cell’s capacity to keep its own internal liquids liquid has been eroded. Understanding and reinforcing that capacity may prove to be one of the most promising directions in the search for effective treatments.

Subject of Research: Regulation of biological phase separation by molecular chaperones and related factors in amyotrophic lateral sclerosis

Article Title: Regulation of biological phase separation by molecular chaperones and related factors in amyotrophic lateral sclerosis

Article References: Regulation of biological phase separation by molecular chaperones and related factors in amyotrophic lateral sclerosis. (n.d.). https://doi.org/10.1186/s12916-026-05261-5

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05261-5

Keywords: amyotrophic lateral sclerosis, phase separation, molecular chaperones, TDP-43, FUS, C9ORF72, nuclear import, proteostasis, protein aggregation, RNA-binding proteins, zinc finger domains, condensates

The research, led by Jiajing Hu and Alfredo Iacoangeli at King’s College London together with Ammar Al-Chalabi, Alexander Schönhuth and a large international network of ALS clinicians, builds on an earlier observation that Capsule Networks, or CapsNets, can learn to distinguish people with ALS from unaffected controls using nothing more than single nucleotide polymorphism genotyping data. CapsNets are a relative newcomer to the deep learning menagerie. Where conventional convolutional neural networks compress spatial information into a single scalar output at each layer, capsules preserve the pose and identity of features as vectors, allowing the network to reason explicitly about part-whole relationships. In genomics, that means the model can, in principle, capture how variants within a gene or pathway combine, rather than treating each single nucleotide polymorphism as an independent bit of evidence. The catch, until now, has been that published CapsNet experiments were confined to specific datasets and genotyping platforms, leaving open whether the approach would generalise beyond the population and the chip it was trained on.

The new work tackles that generalisation problem head-on with a dataset of unusual scale for this disease. The team assembled genetic data from more than 47,000 individuals across 13 countries, drawing on the Project MinE consortium and collaborating ALS centres in Europe, North America, the Middle East and Australia. Crucially, the samples were genotyped on nine different platforms, each of which interrogates a slightly different set of variants and produces its own batch effects, missingness patterns and allele-calling quirks. Any model that silently learns the fingerprint of a particular genotyping chip will fail the moment it meets a patient typed on another one, a failure mode that has plagued machine learning applications in medicine. The researchers therefore engineered their pipeline to be robust to these technical discrepancies, using quality control procedures, principal component analysis to characterise and adjust for ancestry, and a gene-based principal component strategy that compresses variant-level information into gene-level features before the network ever sees it.

That gene-based dimensionality reduction is one of the study’s key technical moves. Rather than feeding hundreds of thousands of raw SNP genotypes into the CapsNet, the authors aggregate variants by gene, computing principal components that summarise each individual’s genetic variation within each gene region. This does two things at once. It shrinks the input space to a size the network can learn from without overfitting, and it anchors the features in biological units, genes, that are shared across genotyping platforms even when the exact SNP content differs. The capsules then operate on these gene-level representations, and the routing-by-agreement mechanism of the CapsNet allows lower-level gene features to vote on the higher-level decision of whether the genome belongs to someone with ALS. The result is a model that is flexible enough to accept an individual external sample, whatever platform produced it, and still produce a calibrated risk estimate.

The performance results are striking for a disease whose common genetic architecture has proved so elusive. In external validation across diverse ALS populations from multiple countries, the model achieved high precision and sensitivity in separating individuals with ALS from unaffected controls, holding up despite the technical and ancestral heterogeneity of the data. The authors are careful to frame this as risk stratification and diagnostic support rather than a stand-alone diagnosis; the model does not replace clinical examination, electromyography or targeted mutation testing. But the ability to extract a disease signal from genome-wide genotyping data in sporadic ALS, where classical genetics has explained so little, suggests that the collective weight of thousands of common variants, and the nonlinear interactions among them, carries real information that linear polygenic risk scores have struggled to capture.

Perhaps the most provocative result comes from a simulation. The team modelled what would happen if the tool were deployed as a population screening test for ALS, and found that its predictive performance under that scenario was comparable to published estimates for screening based on the major known ALS-causing mutations such as FUS and C9orf72. That comparison matters because mutation-based screening only works for carriers of those specific variants, whereas the deep learning model applies to everyone. In effect, the study suggests that a genome-wide pattern learned by a neural network can match the yield of looking for the single most consequential known mutations, at least in the simulated screening context. If that holds up prospectively, it would mean genetic risk assessment could be extended to all individuals regardless of family history or mutation status, a genuine democratisation of ALS genetics.

The clinical logic is compelling because ALS is a disease where time is function. The average diagnostic delay from first symptoms to confirmed diagnosis remains long, often around a year, and there is growing evidence that the earlier treatment begins, the better the outcome. Riluzole, the oldest approved therapy, and the newer antisense and antibody-based approaches now entering the clinic all work best early. A validated genomic risk score could flag individuals at elevated risk for closer monitoring, help neurologists interpret ambiguous early presentations, and, once a diagnosis is made, contribute to prognostic stratification. The authors are explicit that their tool supports, rather than replaces, existing genetic testing: for the minority of patients with known pathogenic variants, targeted testing remains essential, particularly since those variants now determine eligibility for precision therapies such as antisense oligonucleotides.

The study also carries a sober set of caveats that the authors do not shy away from. The model was trained and validated on case-control data, which is the right starting point but not the same as predicting who will develop disease in a healthy population. ALS is rare, so even a highly precise classifier will generate false positives at population scale, and the ethical weight of telling a healthy person they carry elevated risk for a uniformly fatal disease is enormous. Ancestry is another live concern: despite the international scope of the dataset, most participants are of European descent, and the authors stress that further prospective clinical validation in diverse populations is required before any diagnostic use. The team’s own conclusion is measured: the method could support genetic risk stratification now, and diagnostic support in the future, but only after that prospective validation is done.

What makes the work resonate beyond ALS is its demonstration that deep learning can extract clinically meaningful signal from the genomic dark matter of a complex neurodegenerative disease. Capsule Networks were designed for computer vision, and their migration into genomics is part of a broader convergence in which architectures built for one domain prove unexpectedly adept at another. The gene-based principal component trick, the platform-agnostic preprocessing and the multi-country external validation form a template that other common-disease geneticists can copy. For the ALS community, which has watched genetic discoveries accumulate without translating into risk prediction for most patients, the study offers something rarer than a new gene: a working bridge between the genome-wide association era and the clinic. The 47,000 genomes behind the model were donated by patients and families across 13 countries, and the authors dedicate the tool’s promise to exactly that population, the sporadic cases who have, until now, been told their genetics held no answers.

Subject of Research: Deep learning genomic risk prediction for sporadic amyotrophic lateral sclerosis

Article Title: Towards a deep-learning genomic tool for risk stratification and diagnostic support in sporadic ALS

Article References: Hu, J., Pain, O., Al Khleifat, A., Shatunov, A., Andersen, P. M., Başak, N. A., Cooper-Knock, J., Corcia, P., Couratier, P., de Carvalho, M., Drory, V., Gotkine, M., Landers, J. E., Glass, J. D., McLaughlin, R., Pardina, J. S. M., Morrison, K. E., Pinto, S., Povedano, M., … Iacoangeli, A. (2026). Towards a deep-learning genomic tool for risk stratification and diagnostic support in sporadic ALS. Genome Medicine. https://doi.org/10.1186/s13073-026-01744-5

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01744-5

Keywords: ALS, deep learning, Capsule Networks, genomics, risk stratification, sporadic ALS, Project MinE, polygenic risk, genome-wide association study, diagnostic support, C9orf72, Genome Medicine

The research team, led by investigators at Barrow Neurological Institute in Phoenix alongside collaborators at the Translational Genomics Research Institute and City of Hope, tackled a persistent blind spot in dementia research. Most studies of diseased brain tissue grind up whole chunks of cortex, producing an average signal in which the synapse’s voice is drowned out by everything else. To overcome this, the researchers isolated synaptosomes, tiny sealed sacs of presynaptic and postsynaptic machinery that pinch off during gentle tissue disruption and can be purified by centrifugation. From frozen frontal cortex samples of ten patients with confirmed C9ORF72-linked FTD and ten neurologically normal donors, obtained through the Queen Square Brain Bank in London, they harvested these synaptic compartments and subjected them to parallel mass spectrometry proteomics and RNA sequencing.

The technical validation was rigorous. Western blots confirmed that the purified fractions were loaded with presynaptic proteins such as synaptophysin and postsynaptic markers like Homer1, while transmission electron microscopy revealed intact membrane-enclosed terminals studded with synaptic vesicles and electron-dense postsynaptic material. Principal component analyses showed that fraction identity, not random variation, was the dominant source of variance in both the protein and RNA datasets, and nuclear markers were largely absent from the synaptic fractions. In other words, the team had genuinely captured the synapse, not a contaminated soup, giving them a trustworthy window onto synaptic biology in the diseased human brain.

What that window revealed was striking. The proteomic analysis identified 1,324 proteins at altered abundance in C9ORF72-FTD synaptosomes, with 555 increased and 769 decreased relative to controls. Pathway analysis using the synapse-specific SynGO database showed that the downregulated proteins clustered heavily around core neurotransmission machinery: the synaptic vesicle cycle, chemical synaptic transmission, and trans-synaptic signaling. Upregulated proteins, by contrast, were modestly enriched for structural remodeling processes such as synapse assembly and postsynaptic actin organization, a pattern the authors interpret as evidence of widespread architectural rebuilding layered on top of failing communication machinery. Many of the altered proteins were unique to the synaptic fraction, and some even shifted in opposite directions between whole tissue and synapses, hinting at localized redistribution of proteins that whole-tissue studies would completely miss.

The RNA side of the story was just as dramatic. Sequencing of synaptosomal RNA uncovered 2,835 differentially expressed protein-coding genes, including 1,763 upregulated and 1,072 downregulated transcripts. When the researchers compared RNA changes with protein changes in the same synaptic compartments, the correlation was only moderate, with a Pearson coefficient of 0.33. Nearly forty percent of the altered genes showed discordant behavior between transcript and protein levels, a signature of broken post-transcriptional regulation inside diseased synapses. In healthy neurons, local RNA and protein abundance are tightly coordinated to keep synaptic transmission running; in C9ORF72-FTD, that coordination visibly frays.

To test whether these brain-tissue signatures could be modeled in the laboratory, the team differentiated induced pluripotent stem cells from C9ORF72-FTD patients and healthy controls into cortical neurons and matured them for one hundred days in culture. The patient-derived neurons showed clear structural deficits: the density of postsynaptic Homer1 puncta at excitatory synapses was reduced, gephyrin puncta at inhibitory synapses were diminished, and the number of fully colocalized inhibitory synapses dropped significantly, even as the corresponding presynaptic markers remained intact. Longitudinal recordings on multielectrode arrays captured a progressive decline in spontaneous network firing between days 65 and 100, indicating worsening hypoexcitability, although the synchrony of network activity was preserved. Notably, this hypoexcitability contrasts with the hyperexcitability often reported in ALS motor neuron models, underscoring that neuronal subtype and disease phenotype shape how the same genetic mutation plays out at the circuit level.

Critically, synaptosomes isolated from these cultured neurons reproduced key molecular pathways seen in the patient brain. The iPSC-derived synaptic proteome contained 962 differentially abundant proteins, and its RNA profile showed 402 differentially expressed protein-coding genes. When the researchers integrated all four datasets, brain protein, brain RNA, iPSC protein, and iPSC RNA, a set of conserved pathways emerged: synapse organization, synapse assembly, regulation of postsynaptic neurotransmitter receptor levels, and trans-synaptic signaling. Among the consistently downregulated genes were members of the neuronal pentraxin family, NPTX1 and NPTXR, which are known to support excitatory synapse formation and have previously been flagged as synaptic biomarkers in the cerebrospinal fluid of genetic FTD patients.

The most surprising discovery, however, came from a targeted search for cryptic exons. When the RNA-binding protein TDP-43 is lost from the nucleus, a hallmark of nearly all FTD and ALS cases, it stops repressing aberrant splice sites, and strange pseudo-exons get stitched into mature transcripts, often destroying the proteins they encode. Cryptic exons in genes like UNC13A and STMN2 are well-documented markers of this process, but they were assumed to be a nuclear phenomenon. The new data show otherwise. In patient synaptosomes, cryptic exon inclusion in STMN2 was significantly increased, detected in eight of ten C9ORF72-FTD cases, and cryptic transcripts of the synaptic scaffolding gene KALRN were also present in a subset of cases. Strikingly, UNC13A cryptic exons were abundant in whole-tissue homogenates but essentially absent from the synaptic fraction, with STMN2 showing the highest synaptic enrichment ratio. This transcript-specific pattern suggests that some aberrantly spliced RNAs are selectively trafficked or retained at synapses rather than randomly redistributed.

The implications ripple outward. STMN2, also known as stathmin-2, is normally transported to distal neuronal compartments where it supports axonal maintenance and repair; KALRN encodes a scaffolding protein central to synaptic signaling and actin remodeling. If cryptic versions of these RNAs accumulate at synapses, they could disrupt local protein synthesis or RNA homeostasis in ways that directly undermine synaptic function, potentially linking TDP-43 loss of function to the synapse loss observed in living patients through neuroimaging and in postmortem tissue. The authors caution that the presence of these transcripts at the synapse does not yet prove they are locally translated or actively toxic, and future work with spatial transcriptomics and high-resolution imaging will be needed to pin down their exact sub-synaptic location and fate.

Limitations remain, as they do in any study of end-stage human disease. The postmortem synaptosomes represent surviving synapses rather than those already eliminated, so some observed changes may reflect compensatory responses rather than primary causes of degeneration. The iPSC cultures, meanwhile, are simplified neuron monocultures lacking the glial cells that shape synapses in the living brain, and they did not show TDP-43 mislocalization or cryptic exons, highlighting gaps in the model. Even so, the convergence of findings across independent systems lends strong support to a reframed view of C9ORF72-FTD: not simply a disease of dying neurons, but a disease in which the synapse, the brain’s most information-dense structure, becomes a site of RNA-level corruption. If synaptic RNA misprocessing proves to be an early event, it could open a new frontier for therapies aimed at preserving synaptic integrity before the damage becomes irreversible.

Subject of Research: Synaptosome multi-omics of C9ORF72-linked frontotemporal dementia reveals synaptic protein and RNA dysregulation and synaptic localization of TDP-43-associated cryptic exon transcripts

Article Title: Integrative synaptosome multi-omics reveals disrupted synapse organization and localized cryptic transcripts in C9ORF72-frontotemporal dementia

Article References: Spillman, A. M., Alsop, E. B., Gittings, L. M., Garcia-Mansfield, K., Piras, I., Bonfitto, A., Martinez, M. N., Sharma, R., Preller, K. R., Huentelman, M., Pirrotte, P., Van Keuren-Jensen, K., & Sattler, R. (2026). Integrative synaptosome multi-omics reveals disrupted synapse organization and localized cryptic transcripts in C9ORF72-frontotemporal dementia. Acta Neuropathologica, 152(1), Article 41. https://doi.org/10.1007/s00401-026-03084-5

Image Credits: AI Generated

DOI: 10.1007/s00401-026-03084-5

Keywords: frontotemporal dementia, C9ORF72, ALS, synaptosome, TDP-43, cryptic exons, proteomics, transcriptomics, iPSC cortical neurons, synaptic dysfunction, STMN2, UNC13A