A model built from 34 circular RNAs measured in blood predicted progression to symptomatic Alzheimer's disease more accurately than plasma phosphorylated tau-217 or amyloid PET, according to a study published in Nature Medicine on 1 July 2026 by Bridget Phillips, Carlos Cruchaga and colleagues at Washington University School of Medicine in St. Louis. The 34 markers were identified in blood from 1,221 people in the Knight Alzheimer Disease Research Center cohort, then tested in independent samples from a second Knight-ADRC group (551 people: 76 with Alzheimer's disease, 475 cognitively unimpaired) and from the preclinical A4 study (1,767 people, almost all cognitively unimpaired at baseline). For classifying biomarker-confirmed disease status, the circular RNA model reached an area under the curve of 0.945, against 0.877 for plasma p-tau217; combining the two reached 0.977.
The prognostic result is the one the authors lead with. For identifying who would become symptomatic within five years, the circular RNA model reached an AUC of 0.870 against 0.676 for p-tau217 alone (P = 1.86 x 10-5), in an analysis of 78 people who progressed and 610 who remained cognitively unimpaired. In the Knight-ADRC samples the circular RNA model carried a hazard ratio of about 2.9 for progression, compared with about 1.8 for p-tau217, and also outperformed amyloid PET status. The authors report that the combined circular RNA signal begins to diverge from normal roughly two to four years before symptoms appear, and that the model had low predictive power for Parkinson's disease, dementia with Lewy bodies and frontotemporal dementia, which the researchers read as evidence of specificity to Alzheimer's biology.
Circular RNAs are closed loops of non-coding RNA, enriched in brain tissue and resistant to degradation, and they represent a different analyte class from the phosphorylated tau assays now entering clinics. That novelty comes with unfinished business. The samples were cross-sectional and drawn from well characterised research volunteers, not from the mixed, multi-morbid population a memory clinic or a family doctor sees; results were broadly similar across European, African and mixed-ancestry groups, but some of those subgroups were small. There is no regulator-cleared test. Speaking to Alzforum in December 2025 about earlier circular RNA data, Timothy Hohman of Vanderbilt University, a collaborator of the group and an adviser to the company commercialising the technology, noted that the approach needs specialised blood collection tubes that stabilise RNA, something that would have to be built into routine clinical workflow.
Commercial interests sit close to the science. On 2 July 2026, the day after publication, the San Diego company Circular Genomics announced the launch of CircPATH, its circular RNA discovery and validation platform, citing the study; Cruchaga was named to the company's scientific advisory board in a company announcement of December 2025. Predicting the timing of symptoms is also a different question from predicting who benefits from treatment, and the study did not test whether acting on a circular RNA result changes any clinical outcome.