Early subtypes and progressions of progressive supranuclear palsy: a data-driven brain bank study

• Published in: Journal of neurology Vol. 272; no. 10; p. 704
• Main Authors: Ono, Daisuke, Sekiya, Hiroaki, Ghayal, Nikhil B., Maier, Alexia R., Roemer, Shanu F., Uitti, Ryan J., Litvan, Irene, Josephs, Keith A., Wszolek, Zbigniew K., Dickson, Dennis W.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 17.10.2025; Springer Nature B.V
• Subjects: Aged; Aged, 80 and over; Algorithms; Alzheimer's disease; Autopsies; Autopsy; Basal ganglia; Brain - pathology; Central nervous system diseases; Cerebrovascular disease; Chi-square test; Clinical trials; Clustering; Decision trees; Disease Progression; Female; Humans; Language disorders; Large language models; Male; Medical prognosis; Medicine; Medicine & Public Health; Middle Aged; Movement disorders; Neurology; Neuropathology; Neuroradiology; Neurosciences; Original Communication; Pathology; Patients; Phenotypes; Posture; Progressive supranuclear palsy; Python; Speech; Supranuclear Palsy, Progressive - classification; Supranuclear Palsy, Progressive - pathology; Supranuclear Palsy, Progressive - physiopathology; Survival analysis; Tau protein; Tissue Banks; Florida; United States > US; Tauopathy; Progressive supranuclear palsy; ChatGPT; Artificial intelligence; Machine learning
• ISSN: 0340-5354; 1432-1459; 1432-1459
• DOI: 10.1007/s00415-025-13459-5

Background Progressive supranuclear palsy (PSP) is typically characterized by vertical supranuclear gaze palsy and early falls, referred to as Richardson’s syndrome (PSP-RS). Other presentations include postural instability (PSP-PI), Parkinsonism (PSP-P), speech/language disorder (PSP-SL), frontal presentation (PSP-F), ocular motor dysfunction (PSP-OM), and corticobasal syndrome (PSP-CBS). However, differences across the early presentations and their subsequent clinical courses remain to be elucidated. Objective This study aimed to characterize early PSP subtypes and their subsequent progressions using a large postmortem dataset. Methods An automated pipeline incorporating fine-tuned Chat Generative Pretrained Transformer (ChatGPT) was developed. The pipeline collected 195 clinical features with onset information from autopsy-confirmed PSP cases without significant neurodegenerative co-pathologies. Results A structured clinicopathologic dataset from 588 patients was analyzed. The results from unsupervised clustering were distilled into a decision tree model. The mutually exclusive algorithm identified seven subtypes: PSP-PF (postural and frontal dysfunction), PSP-RS, PSP-PI, PSP-P, PSP-SL, PSP-F, and PSP-OM, based on five clinical manifestations: frontal presentation, postural instability, ocular motor dysfunction, speech/language disorder, and Parkinsonism. PSP-PF showed rapid progression, the shortest median disease duration (six years), and high tau burden in cortical and subcortical regions. In PSP-F, frontal presentation preceded other symptoms by four years, with a nine-year disease duration—second longest after PSP-P (10 years). PSP-CBS was not identified as an independent subtype. Conclusions This data-driven study identified a novel, aggressive PSP phenotype characterized by early postural and frontal dysfunction. Early subtyping using the decision tree model would help clinicians estimate progression and facilitate early patient recruitment for clinical trials.