Splice-site variants in neurology: from molecular mechanisms to clinical interpretation - a focused review.

Focused review of variants disrupting pre-mRNA splicing in Mendelian neurological disorders, from deep learning-based prediction (SpliceAI, Pangolin) to RNA-level validation by RNA sequencing, long-read transcriptomics and minigene assays

Variants disrupting pre-mRNA splicing are a historically underrecognized share of pathogenic alleles in Mendelian neurological disorders, estimated at 15 to 30% of disease-causing variants, and clinical pipelines frequently overlook them because analysis and reporting focus on protein-coding regions. This focused review addresses the practical problem of how a splice-altering variant found on a diagnostic report can be reliably identified, experimentally validated and translated into a clinical diagnosis or therapeutic decision. Deep learning-based splice predictors such as SpliceAI and Pangolin, RNA sequencing, long-read transcriptomics and minigene functional assays have collectively improved diagnostic yield in previously unsolved neurogenetic cohorts, notably by identifying deep intronic pseudoexon variants and leaky canonical splice-site alleles. Classification remains challenging: the PVS1 criterion of the ACMG/AMP framework requires gene- and variant-specific adaptation when applied to splicing, prediction tools disagree with one another and are least reliable outside the core splice-site consensus, and RNA-based validation is constrained by the tissue- and developmental-stage-restricted nature of splicing. The review covers epilepsy, hereditary neuropathy, ataxia, dystonia, mitochondrial disease, neuromuscular disorders and the newly recognized neurodevelopmental disorders linked to spliceosomal small nuclear RNAs (snRNAs), and emphasizes the priorities for embedding RNA-level diagnostics more routinely into clinical neurogenetic practice.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

The merit of this review is to ask the laboratory's question — how to get from a predicted splice variant to a diagnostic decision — rather than whether splicing matters, which is settled, and the three bottlenecks it names (adapting PVS1, disagreement between predictors, RNA validation limited by tissue- and developmental-stage-restricted splicing) are those of real-world reporting. The abstract announces, however, no systematic literature search, no number of included studies and no operational decision tree: the priorities for embedding RNA into diagnosis are stated, not turned into a course of action. The 15 to 30% estimate of causal variants should be read as an order of magnitude from a focused review, not as a measurement.

Analysis by Dr Thibaut Benquey