Blood Biomarkers Advance Early Alzheimer's Detection With New Diagnostic Capabilities

Blood-based biomarkers for Alzheimer's disease are advancing diagnosis and patient care by facilitating earlier disease confirmation, though their use raises new questions about appropriate testing contexts and the need for confirmatory procedures. The new tests can help narrow down possibilities when a patient starts showing signs of cognitive decline, but they are not yet at a point to make a definitive diagnosis. Before a patient can start treatments that are now available to remove amyloid plaques, they still need an amyloid PET scan or cerebrospinal fluid evaluation to confirm what their blood test shows.

The identification and validation of biomarkers for neurodegenerative diseases is advancing rapidly, playing a critical role in enabling early diagnosis, tracking disease progression, and assessing how patients respond to treatment. With the recent approval of monoclonal antibody therapies such as lecanemab (tradename: Leqembi®) and donanemab (Kisunla™), the clinical imperative for accurate preclinical and early-stage AD diagnosis has intensified. Traditional diagnostic modalities - cerebrospinal fluid neurochemical dementia diagnostics and amyloid-PET imaging - are effective but impractical for large-scale screening due to invasiveness, cost, and logistical constraints.

Plasma brain-derived tau (BD-Tau) represents a novel biomarker designed to overcome limitations of traditional total tau measurements. The majority of blood total tau originates from peripheral tissues, not the CNS, and is elevated in a range of conditions, including Creutzfeldt-Jakob disease, head trauma, and anoxia, among others. This lack of specificity has led to its removal from the revised AT(N) criteria for AD diagnosis.

To address this, researchers developed a tau junction antibody (tau J), which specifically binds the junction between exon 4 and 5 – unique to CNS tau. This antibody does not bind if the "big tau" insert is present, thus excluding peripheral tau from detection. The antibody was tested using ELISA and immunoprecipitation-mass spectrometry, and an immunoassay was developed on the Simoa platform.

Initial studies using neuropathologically confirmed cohorts demonstrated that plasma BD-Tau levels increase in the presence of AD pathology, in contrast to total tau, which remains unchanged across the AD continuum. Memory clinic cohorts from Italy further confirmed that plasma BD-Tau is significantly elevated in AD compared to other neurological conditions, mirroring the specificity of CSF total tau. Unlike neurofilament light chain, which is elevated in various neurodegenerative diseases, BD-Tau shows greater specificity for AD.

Research in collaboration with Norwegian cohorts revealed that plasma BD-Tau not only distinguishes AD from other conditions but also predicts longitudinal cognitive decline and changes in MRI meta-ROI signatures. Importantly, patients who are double positive for amyloid (phospho-tau 217 or 181) and BD-Tau in plasma exhibit the fastest rates of cognitive and structural decline, suggesting that BD-Tau adds prognostic value beyond amyloid markers alone.

The utility of BD-Tau extends beyond chronic neurodegeneration. In traumatic brain injury, plasma BD-Tau remains elevated for up to seven days post-injury, distinguishing patients with poor outcomes from those with better recovery. In acute ischemic stroke, BD-Tau measured at admission is the only marker capable of differentiating patients with good versus poor 90-day functional outcomes. The correlation between plasma BD-Tau and stroke lesion size is strong and independent of vascular territory or anatomical location.

A key technological advance in blood-based biomarker detection is the use of immunoprecipitation as a preparatory step before high-sensitivity immunoassays, collectively termed IP-IA. This approach enhances the detection of low-abundance biomarkers and improves diagnostic accuracy by enriching target proteins and clearing interfering matrix components. The IP-IA methodology combines robotic IP with high-throughput immunoassays. The initial IP step uses functionalized magnetic beads with antibodies targeting diagnostic proteins (Aβ peptides, tau proteins), enriching the biomarkers and removing matrix components that can mask epitopes or interfere with detection.

Following IP, eluates are analyzed using commercial immunoassay platforms (e.g., ELISA, MesoScale, Simoa, Lumipulse, Elecsys), enabling multiplexed detection of Aβ and tau epitopes from minimal plasma volumes (as little as 100 microliters). Multiple studies have demonstrated that IP-IA significantly improves the diagnostic accuracy of blood-based Aβ and tau measurements compared to direct assays. In collaboration with Roche Diagnostics, prior IP increased the area under the curve for Aβ42/Aβ40 ratio detection from 0.73 to 0.88, and further to 0.92 with biomarker-supported clinical diagnosis.

The earliest detectable biomarker change in AD is a decrease in CSF Aβ42, specifically the Aβ42/Aβ40 ratio. This change is mirrored in blood, occurring concurrently with CSF alterations. Other biomarkers, such as phospho-tau isoforms (notably p-tau217 and p-tau181), rise later in the disease course. Importantly, some patients with subjective cognitive deficits or early MCI may not show pathological amyloid-PET or tau-PET findings, underscoring the need for sensitive blood-based assays.

Researchers at Vellore Institute of Technology, Tamil Nadu combined interpretable deep learning with blood gene expression to offer a scalable, non-invasive approach for early Alzheimer's detection. They combined three public microarray datasets — GSE63060, GSE63061, and ADNI — into a single dataset of 476 samples with 12,459 common genes. To find the most informative genes, the team applied multiple feature selection methods, including Chi-square, ANOVA, Recursive Feature Elimination, and ElasticNet. They then used SHapley Additive exPlanations (SHAP), an explainable AI approach, to rank genes by importance and improve interpretability.

Two deep learning models — a deep neural network (DNN) and a 1-D convolutional neural network — were trained on the selected genes. The DNN reached 91% accuracy and 95% precision, outperforming conventional machine-learning methods. Among the genes identified, RPL36AL, CSF2RB, and RMND5B emerged as key biomarkers, highlighting molecular signatures that distinguish Alzheimer's patients from healthy individuals.

The advent of blood tests raises questions about appropriate use. People who are completely normal but who may have a family history of Alzheimer's disease or are just anxious and want a test may manage to get one even though it's not appropriate for them, and then if they get a false-positive result they may struggle with what that means for them. Alzheimer's disease is a slow process that happens in the body over a period of time. The estimate now is that some of the changes may be going on for 15 or 20 years before a person has clinical symptoms. There is a phase called preclinical Alzheimer's disease where a person is completely normal cognitively. If such a person were to do these blood tests, the result would show up as positive. But it is not known if that person will develop symptoms or may not develop symptoms. And if they do develop symptoms, it is not known if it will be in the next five, 10 or 15 years. That's why there's a need to only do testing when there's clearly documented cognitive impairment.