Personal CSF Protein Baselines Can Mask Alzheimer’s Biomarkers

Doctors are increasingly using proteins in spinal fluid to support the diagnosis of Alzheimer’s disease and other brain disorders. Our research shows that each person has a characteristic background level of brain-derived proteins in their cerebrospinal fluid, a personal “baseline signature”. This hidden signature can blur the signal from disease. The good news is that we can adjust for it.

In a study published in Scientific Reports, our team examined spinal fluid from 499 people with different neurodegenerative diseases, including Alzheimer’s disease, several forms of frontotemporal dementia, corticobasal syndrome, progressive supranuclear palsy, amyotrophic lateral sclerosis (ALS), and a group of healthy volunteers. One of our main questions was: how much of the variation we see in spinal fluid proteins is caused by disease, and how much is simply because people differ from one another at baseline?

Cerebrospinal fluid (CSF) bathes the brain and spinal cord. It is produced inside the brain, circulates through the ventricles and around the brain, and is then cleared back into the blood. Because it is in such close contact with brain tissue, it contains many proteins that come from nerve cells and other brain cells.

In the clinic, doctors collect CSF with a lumbar puncture (also called a spinal tap), where a thin needle is inserted in the lower back to draw a small amount of fluid. Today, the most widely used CSF tests for Alzheimer’s disease measure different forms of two key proteins, beta-amyloid and tau. The 42-amino-acid form of beta-amyloid (often called Aβ42) usually decreases in CSF when amyloid plaques build up in the brain. Certain phosphorylated forms of tau (p-tau) instead rise when neurons are injured, and tau tangles develop.

These markers are examples of protein tests that have transformed research and clinical trials, and they are now widely used in specialized memory clinics. Still, only a handful of proteins suggested to have clinical relevance have made it into routine care as many promising biomarkers have turned out to be less reliable than hoped. One possible reason is that people differ a lot in their overall CSF protein levels, even when they are healthy.

A personal background level of brain proteins

We measured 69 brain-related proteins in the CSF of 499 individuals. To do this, we used a technology called suspension bead arrays, from Luminex, which allows us to measure many different proteins at the same time, using tiny, color-coded beads and only very small amounts of spinal fluid.

Here, the Human Protein Atlas (HPA) project based at SciLifeLab, plays a central role. Over many years, this Swedish initiative has built a public, open-access map of where each human protein is found in the body, using carefully validated antibodies and other methods. As part of this effort, the project has generated a large collection of so-called HPA antibodies that are manufactured in Sweden by Atlas Antibodies. These antibodies are designed to recognize specific human proteins and the catalogue, now comprising of more than 23,000 antibodies, covers over 75% of the human proteome.

In our work, we coupled selected HPA antibodies to microscopic, color-coded beads and used them as “hooks” to capture their target proteins from spinal fluid. We then measured the signal from each bead to estimate the levels of each protein.

When we analyzed the data, the first thing that stood out was not the difference between diseases and healthy volunteers. Instead, it was the difference between individuals.

Many of the proteins that mainly originate from the brain moved together. Some people had low levels across most of these proteins, others had high levels across the board. This personal background level explained more than 70% of the variability in the data.

This overall level was not strongly linked to sex, and only weakly related to age or diagnosis. In other words, long before we look at disease, each person seems to have a characteristic overall level of brain-derived proteins in their spinal fluid. This baseline is possibly shaped by factors such as brain volume, CSF production and clearance, and other aspects of physiology that we still do not fully understand.

From a diagnostic perspective, this personal background level is a problem if we ignore it. Many disease-related changes are relatively modest compared with the variation between people. If we simply compare raw protein levels between patients and controls, we risk missing true disease signals in someone who starts very high or very low at baseline.

To test this idea, we built statistical models that effectively adjusted each person’s protein levels by their own median level of the “brain-derived protein cluster”. Then we asked how well individual proteins separated each disease group from healthy controls.

When we adjusted for this personal background, the association between many proteins and disease became stronger, and their ability to distinguish patients from healthy individuals improved. Some proteins that looked unremarkable at first only became clearly different after this correction, especially those that were decreased in disease.

However, most of these proteins were still not specific for a single disease. Many showed similar changes across several neurodegenerative conditions, possibly reflecting shared processes such as loss of synapses (the contact points between neurons) and damage to axons (the long nerve fibers that carry signals).

Protein pairs as a practical solution

Adjusting for a person’s overall brain protein level works well in research, where many proteins are measured at once. But clinical laboratories rarely measure dozens of proteins in a single run; they may focus on just one or a few biomarkers.

We therefore asked whether a simpler trick could capture some of the same information. The idea is familiar from existing practice. Clinicians often use the ratio between Aβ42 and Aβ40 rather than Aβ42 on its own. The shorter form, Aβ40, acts as an internal reference that reflects how much amyloid a person produces in general.

We extended this reasoning to other proteins. For each disease, we identified several proteins that tended to increase and several that tended to decrease after proper adjustment. Then we combined them into pairs in which one protein went up in disease and the other went down.

Autoimmunity: When Immune Defense Turns Inward

This infographic explores how the immune system fails, diagnostic methods used to detect autoimmune diseases, and emerging therapeutic innovations from CAR-Tregs to bispecific antibodies.
View Infographic

Advertisement

When we did this for all possible pairs drawn from the best candidates, many pairs outperformed the individual proteins.

One instructive example in Alzheimer’s disease was a pair consisting of growth associated protein 43 (GAP43) and protein tyrosine phosphatase receptor type N2 (PTPRN2). Each of these proteins is informative on their own once adjusted for background. Together, they produced a very strong separation between patients and healthy controls in our dataset, because they pull in opposite directions and partly correct for each other.

Rethinking some puzzling biomarker profiles

Our findings also shed light on a group of patients that has been difficult to interpret in recent years. These are individuals whose CSF profile is amyloid-negative but tau-positive, when judged by standard cut-offs for the Aβ42/Aβ40 ratio and phosphorylated tau.

This pattern has raised concern, because it does not fit neatly into current biological definitions of Alzheimer’s disease. In our study, individuals with this profile tended to have higher overall levels of brain-derived proteins in CSF.

This suggests that at least some of them might not have true tau pathology, but rather globally increased protein concentrations caused by altered CSF circulation or other physiological factors.

To advance research into translation, we need several things. We need larger multi-center studies that include many diseases in the same analysis, rather than only one. We need longitudinal sampling to see how protein patterns change over time within individuals. We need links to imaging and genetics, to learn which aspects of anatomy and physiology drive the personal background level. We also need robust validation of the most promising protein pairs in independent cohorts.

Neurodegenerative diseases already place a heavy burden on patients, families, and healthcare systems. We will not meet that challenge with a single “magic” biomarker. But by combining better biological insight and careful statistical thinking, we can move steadily closer to spinal fluid and blood tests that are both accurate and clinically meaningful.

Reference: Mravinacová S, Bergström S, Olofsson J, et al. Addressing inter individual variability in CSF levels of brain derived proteins across neurodegenerative diseases. Sci Rep. 2025;15(1):668. doi: 10.1038/s41598-024-83281-y