Researchers at Boston University applied subtype and stage inference (SuStaIn), an unsupervised machine learning method, to quantitative hyperphosphorylated tau (pTau) density measured in 26 brain regions from 207 donors with autopsy-confirmed chronic traumatic encephalopathy (CTE) and 75 control donors, and report three distinct patterns of tau progression rather than a single one. The paper, by Fatima Tuz-Zahra, Jonathan Cherry, Thor Stein, Ann McKee and colleagues, was published in Acta Neuropathologica on 23 June 2026 (volume 151, article 69). All donors came from the university's Understanding Neurologic Injury and Traumatic Encephalopathy (UNITE) Brain Bank, and neuropathologic assessment was done blinded to clinical information. Donors meeting criteria for Alzheimer's disease, frontotemporal lobar degeneration, motor neuron disease or neocortical Lewy body disease were excluded.

The algorithm labelled 93 cases Subtype 1, described as cortical predominant, with dense and rapidly progressing cortical pTau resembling classically described CTE; 105 cases Subtype 2, described as cortical sparing and hippocampal enhanced, with less cortical tau, slower progression and pronounced hippocampal involvement in a CTE-specific pattern; and 28 cases Subtype 3, described as copathology altered, older donors at higher CTE stage enriched for phosphorylated TDP-43, amyloid beta, hippocampal sclerosis and arteriolosclerosis. Clinically, scores on the Functional Activities Questionnaire, an informant report of everyday tasks such as paying bills or keeping appointments, correlated with regional tau burden only in Subtype 1 cases. The authors write that Subtypes 1 and 2 might represent two neuropathologically and clinically distinct forms of CTE.

The finding sits against a diagnostic framework built on a single ascending scale. CTE can still only be confirmed after death, on the pathognomonic lesion defined at the second NINDS/NIBIB consensus meeting in 2021, and severity is conventionally described along one linear staging axis. A cortical sparing variant was described separately in Acta Neuropathologica in February 2024 and is not part of consensus criteria; the new analysis reports that donors meeting that description cluster in its Subtype 2, and that the ratio of frontal to hippocampal tau was significantly higher in Subtype 1 than Subtype 2, an effect that persisted after adjustment for age at death.

The authors set out the counterweights themselves. The data are cross-sectional autopsy measurements, so the ordering is inferred, not observed. Subtype 3 donors were on average about 15 years older than those in the other two groups, so the copathology pattern could be partly an age effect, and the paper notes that donors in Subtypes 1 and 2 might converge on Subtype 3 given enough time. The authors call for a larger, higher-powered study with more diverse comorbid pathology, and for memory-specific testing to check whether the medial temporal pattern of Subtype 2 tracks a different clinical picture.