The Expanding Role of Blood-Based Biomarkers in Alzheimer Disease

As blood-based biomarkers continue to reshape the landscape of Alzheimer disease (AD) diagnosis and research, advances in highly sensitive detection technologies are enabling researchers to investigate a broader range of disease-related pathologies. Among these emerging markers is extracellular mid-region tau (eMTBR-tau), a fragment of tau associated with the seeding and formation of neurofibrillary tangles that may provide complementary information to established biomarkers such as phosphorylated tau.

At the 2026 Alzheimer's Association International Conference (AAIC) in London, NeurologyLive^® spoke with Stephen Williams, PhD, chief scientific officer at Alamar Biosciences, about the company's approach to biomarker development and the evolving role of blood-based testing in neurologic disease. The conversation touched on the development of eMTBR-tau testing, the potential of multiplex approaches to characterize multiple disease pathologies, and the challenges of translating increasingly complex biomarker data into clinically useful tools.

Williams discussed the technology behind Alamar's platform and how advances in assay sensitivity and specificity have supported the detection of low-abundance biomarkers associated with neurologic disease. He also explored the potential of eMTBR-tau to provide insight into tau tangle pathology, the value of assessing coexisting pathologies such as amyloid, tau, neuroinflammation, and cerebrovascular disease, and how machine learning-based models could eventually help clinicians better understand the factors contributing to an individual patient's cognitive symptoms.

NeurologyLive: For those who may not be familiar with your research, could you provide a general overview of your work?

Stephen Williams, PhD: Alamar Biosciences has developed a platform that occupies a unique space in the biomarker field. It can span the continuum from exquisitely sensitive and specific assays that measure individual biomarkers one at a time to the simultaneous measurement of hundreds of biomarkers for research purposes.

That ability allows researchers to approach biomarker development from different directions. Some start by measuring a large number of biomarkers and then narrow their panels to those that are most predictive of cognitive decline or the presence of cerebrovascular disease. From the other end, you can build what I like to call "pathology sentinels." These are biomarkers with the most provenance that are important and actionable and have often been developed individually through years of intensive research. We can measure them one at a time or aggregate them into actionable subsets of markers.

The ability to do that in a fully automated way, while maintaining the analytical performance, sensitivity, and specificity needed to detect different variants of biomarkers, is the space that we occupy.

Can you explain some of the biggest milestones in the development of this technology and the main takeaways researchers have gained from using it in practice?

I think it started with the concept that if you can suppress background noise in your measurements, you can detect biomarkers or diagnostics that are present at much lower abundances and concentrations. That's where the engineering team started about 10 years ago. They developed a unique chemistry that enabled them to do that, but they weren't satisfied there. They wanted to put that technology into a fully automated system that would make it easier for researchers around the world to use and, ultimately, easier to translate into medical practice.

The match between neurologic disease and this technology is particularly attractive because many of the most important neurologic biomarkers are present at very low concentrations in the blood. They're low abundance because they're produced in the brain, and relatively few of them cross the blood-brain barrier. Many of the markers we're interested in are also subtle variants of proteins, such as phosphorylated tau or specific splice variants. That requires a very high level of sensitivity and an exquisite level of specificity.

That's why we've seen such rapid uptake in neurodegeneration. The field already understood that these capabilities were needed. At the same time, the standards used to establish the presence of disease pathology have evolved over the past 20 years. When developing a blood-based biomarker, we typically want to measure it against a known truth standard. In neurology, that evolution has moved from imaging-based standards, such as amyloid accumulation or the spatial distribution of tau in the brain, to cerebrospinal fluid and now to blood. That progression is making clinical diagnosis much more practical for the future.

What does the ability to measure extracellular mid-region tau (eMTBR-tau) in blood mean for the Alzheimer disease community?

I talked earlier about the concept of pathology sentinels. These are markers that guard or represent a particular physiologic pathway, and eMTBR-tau is the latest piece to fill an important gap.

One of the key opinion leaders described it to me as the "fluffy bits" of tau tangles that fall off and show that you actually have them in the brain. Amyloid has limitations because people can have amyloid in the brain without having Alzheimer disease. To diagnose Alzheimer disease, you need both plaques and tangles.

As good as phosphorylated tau 217 is at detecting amyloid pathology and staging very earlydisease, it doesn't fully capture the mid-to-late stages of the disease process. For physicians, the question becomes: Does this patient who has amyloid pathology also have tau tangles? That's the missing piece that eMTBR-tau may provide.

eMTBR-tau is a specific fragment of tau that represents the causal, seeding form of tau associated with the development of tangles in the brain. Being able to measure that in blood could provide important information about the presence and progression of tau pathology.

Where do you see biomarker testing going from here?

I think it will become increasingly multidimensional. The old idea that we could measure one "magic marker" and have it tell us everything has already become impractical. Even measuring one or two markers isn't going to be enough for every clinical question.

Beyond amyloid and tau, what else would we want to measure? Neuroinflammation, for example, because there may be interventions that can reduce inflammation. We also need to understand cerebrovascular disease. If a patient has cognitive impairment, they may have amyloid, Lewy body pathology, or tau pathology, but it may actually be cerebrovascular disease that is the dominant contributor to their cognitive impairment.

We need to understand the balance of these different pathologies in each individual. We have beautiful graphs at the population level showing when different biomarkers emerge, but those patterns don't necessarily tell us what is happening in an individual patient.For a clinician treating an individual, the question is whether to address cerebrovascular disease, deplete amyloid, or, in the future, deplete tau or reduce neuroinflammation. To make those decisions, clinicians need to understand the different coexisting pathologies that may be contributing to a patient's symptoms.

I think that's where we will see a tension between measuring as few things as possible and reducing the information to a simple yes-or-no answer. We've seen that with amyloid and tau: Is amyloid present or not? Is tau present or not? But I don't think that's enough.

The future will lead us into multiplex testing, where we're measuring multiple things at once. Some of those will continue to be our pathology sentinel markers, while others will be multivariate models derived using artificial intelligence or machine learning.

We've already seen some of these approaches. There are multivariate models that predict cerebrovascular disease, as well as others that predict the presence of TDP-43 pathology. At the moment, there may not be an intervention that can improve TDP-43 pathology, but the tools to measure these pathologies need to come first. You can't develop an intervention for something that can only be measured in someone's brain after you've taken it out of their head.

I think we'll continue to see this evolution. The aggregation of pathology sentinels is the first step into multiplex testing, but we'll also see machine learning-derived models that assess co-pathology, predict cognitive decline, or identify cerebrovascular disease integrated into the diagnostic process.

What will be important to ensure these increasingly complex biomarker tools are practical for clinicians and patients?

I think the technology also has to be made practical. Even if you develop machine learning models that incorporate dozens of proteins, you ultimately have to help the physician make a decision for the individual patient in front of them.

Researchers often love complexity and individual markers, but when you move into the clinic, the information needs to be practical, simple, and cost-effective. I think we can accomplish that because the output of these models doesn't have to be a list of 27 different proteins. It could instead tell a clinician which pathology burden the patient is carrying that is most likely to account for their symptoms.

That's ultimately what I think we'll see. It doesn't mean clinicians have to be separated from the underlying science or the detailed information. I think we can have both: the complexity and depth of the underlying data, combined with a practical output that supports personalized care.

Transcript edited for clarity. Click here for more AAIC 2026 coverage.