February 19, 2026 | Researchers from the Foundation for the National Institutes of Health (FNIH), Washington University in St. Louis, the University of California San Francisco, the University of Wisconsin-Madison, and more have published results of a study using a single plasma biomarker sample to estimate not only the probability of an Alzheimer’s Disease diagnosis, but when symptoms will develop. The work was published today in Nature Medicine.
The results came from the FNIH Biomarkers Consortium Plasma Aβ and Phosphorylated Tau as Predictors of Amyloid and Tau Positivity in Alzheimer’s Disease Project Team and is representative of the important work FNIH does to bring together stakeholders and ideas to accelerate drug discovery and, ultimately, help patients, explained Alessio Travaglia, Director of Translational Science and Neuroscience at the FNIH.
Travaglia views the latest paper as evidence of FNIH’s long commitment to Alzheimer’s research. “Until 20 years ago, we relied on clinical diagnosis to know whether people had Alzheimer’s Disease or not. About 10 years ago we started having the PET imaging scans to understand whether there were changes in the brain,” he said. “What’s happened in the last several years is remarkable and the speed of change is incredible. About a couple of years ago, we had the first blood-based biomarker that could tell whether people have the same biomarkers or changes in the brain measured by the more invasive and expensive procedures.”
The current research takes biomarkers one step further, to predict not just “if” but “when” symptomatic Alzheimer’s Disease will begin.
Amyloid plaques have been shown to build in up an Alzheimer’s brain for 10-20 years before any neurological symptoms are evident; tau neurofibrillary tangles develop later and increase with symptom severity. Identifying and intervening during this preclincal phase—before significant neurodegeneration—is a major goal of drug development efforts.
The researchers sought to develop a mathematical clock model to predict the age of onset of AD symptoms. Clock models create trajectories based on a reference point (amyloid or tau PET positivity) instead of making predictions based on age alone. “Unlike general biological aging clocks or categorical staging based on multiple biomarkers, clock models track disease progression with a specific biomarker,” the authors write.
As a biomarker, the researchers chose the ratio of phosphorylated to non-phosphorylated plasma tau at position 217—written %p-tau217. Phosphorylated tau is not the only biomarker associated with Alzheimer’s progression, but the authors write that plasma measures of phosphorylated tau and %p-tau217, have high associations not only with PET scans looking at amyloid plaque buildup and tau tangle buildup, but also brain volume and cognition scores. The authors write: “%p-tau217 captures the co-evolution of amyloid and tau pathologies across the AD continuum.”
There are also commercial assays available to test for phosphorylated tau and %p-tau217: C2N Diagnostics’ PrecivityAD2 test includes the ratio of phosphorylated to non-phosphorylated tau. Other clinical assays test for phosphorylated tau including LucentAD from Janssen and Quanterix, Simoa ALZpath from Quanterix, and Lumipulse from Fujirebio.
The study looked at cohorts from two longitudinal patients datasets: the Knight’s Alzheimer’s Disease Research Center and the Alzheimer’s Disease Neuroimaging Initiative (ADNI). In total, they considered about 900 patients with at least two plasma %p-tau217 samples from every participant over the course of 5 to 7 years.
Using longitudinal plasma %p-tau217 data, the researchers created clock models that took a current %p-tau217 value and looked back to estimate the point at which a patient became %p-tau217 positive and then estimated forward to predict when symptoms would arise. Two mathematical modeling approaches were used: temporal integration of rate accumulation (TIRA) and sampled iterative local approximation (SILA).
At the upper and lower levels of %p-tau217, predictions were not reliable. “Longitudinal data from individuals with high %p-tau217 values was sparse, providing less certain clock estimates for higher values,” the authors wrote. At very low levels, the TIRA model provided “highly unstable” estimates.
But for the bulk of the data, the two models were consistent across both populations. “Overall, the clock models estimated similar ages at plasma %p-tau217 positivity regardless of the [mathematical] method or [patient] cohort and were consistent with observed conversion ages from %p-tau217 negative to positive,” the authors wrote.
The code developed by the authors for this study is available for download from Github: https://github.com/WashUFluidBiomarkers/plasma_ptau217_time. The research team also developed a web-based application that allows scientists to visualize how levels of plasma p-tau217 change over time and how they relate to Alzheimer’s symptoms. The interactive tool helps researchers to explore complex relationships between plasma p-tau217, age, and symptoms.
The clock model revealed some surprises. First, the effects of APOE4 carrier status, sex, and years of education had minimal impact.
Second, %p-tau217 as a single biomarker “likely captures the intertwined progression of both amyloid and tau pathology,” the authors wrote. Adding other biomarkers in the future— eMTBR-tau243 or biomarkers of cerebrovascular disease—may enable greater precison in estimating time to symptom onset, they wrote.
What surprised Travaglia the most, though, was the age-linked difference in time between %p-tau217 positivity and onset of symptoms—a difference that was consistent across both patient cohorts and both mathematical approaches used to create the clock models. Individuals who became %p-tau217 positive at age 60 were estimated to develop symptomatic Alzheimer’s Disease after about 14 years, while 80 year olds became symptomatic in only 6.2 years. The authors theorized that the difference may come from co-pathologies as patients age, which, “may further complicate the interpretation of %p-tau217 in older individuals.”
“This is something that I didn't necessarily expect,” Travaglia said. “Certainly it's going to help the design of clinical trials moving forward, because I could set a very specific enrollment criteria. I could have a specific age as my inclusion criteria.”
Travaglia emphasized that the clock model is not useful for an individual patient’s decision-making. The model has a median average error range of three to five years and has not been tested on all representative populations. For example, participants in these cohorts were largely non-Hispanic White, which limits the generalizability of the mode. But he still sees strong promise for the model in clinical research settings.
“Let's say I'm a pharmaceutical company and I'm putting together a clinical trial. How can I have the highest chance to design the trial in a way that is going to be successful? What are the people that I should recruit in my trial? What exclusion criteria should [we use]?” explained Travaglia. The clock model will allow trial sponsors to better recruit appropriate patients to get useful data from the trial.
There’s more work to be done, and Travaglia foresees expanding the patient cohorts. “If you want to generalize the finding, we have to look into a larger cohort or bring together multiple cohorts,” he said. This is something that [FNIH] is doing right now with the next directional follow-up here with this study called AD Biosignature Project.”
The rate of progress in Alzheimer’s Disease research is truly astounding, Travaglia said, and he’s thrilled with the team science approach that FNIH is enabling. “It is huge progress; I can't wait for what's coming next because every few months we're going to get something quite interesting in this space.”