Proteomic biomarkers of cognitive function, APOE ε4 status, and dementia in Generation Scotland

Hannah M. Smith, Joanna E. Moodie, G Davies, Anne Richmond, Janet Robertson, Aleksandra D. Chybowska, Camilla Drake, Poppy Adkin, Spyros I. Vernardis, Arturas Grauslys, Matthew E.h. White, Sergej Andrejev, Charles Brigden, Christoph B. Messner, Caroline Hayward, Aleksej Zelezniak, Markus Ralser, Ilse Krätschmer, Matthew R. Robinson, S. R. Cox, R. MarionimedRxiv 2026 · DOI

This study, recently published as a preprint from the Generation Scotland cohort, leverages untargeted mass spectrometry to identify blood-based protein signatures associated with cognitive performance, the APOE ε4 genetic risk factor, and long-term dementia risk. By analyzing over 14,000 individuals, the researchers identified key proteins—including Selenoprotein P, Afamin, and Complement factors—that could serve as early-warning biomarkers for neurodegeneration.

TL;DR

Researchers used high-throughput, untargeted mass spectrometry on a massive Scottish cohort to find protein "fingerprints" in the blood that correlate with brain health. They identified specific proteins linked to cognitive test scores and genetic risk, including a complement protein group that predicts a lower risk of developing dementia up to 17 years before diagnosis.

What Problem Does This Paper Address?

Dementia prevalence is skyrocketing globally, yet we still struggle with early detection. While blood-based biomarkers are the "holy grail" of clinical diagnostics due to their accessibility, most current research relies on affinity-based approaches (like Olink or SomaScan). These methods use pre-defined panels of antibodies to look for specific proteins, potentially missing novel or unexpected biological signals.

This paper addresses the gap by using untargeted mass spectrometry (MS). Instead of looking for what they expect to find, the authors measured 439 independent protein signals to see which ones naturally correlate with cognitive function and disease risk. They also aimed to distinguish between proteins that reflect a person's current cognitive state versus those that reflect lifelong, inherited genetic predispositions.

Methods

The study utilized a sophisticated statistical pipeline to handle the high-dimensional nature of proteomic data:

  • Untargeted Proteomics: They used ultra-high-throughput liquid chromatography-mass spectrometry (specifically the SWATH-MS method) to measure serum protein levels in over 14,000 participants.
  • MAJA (Multivariate Bayesian Regression): To analyze cognitive function, they employed a novel "joint sparse" model. Unlike standard regressions that look at one outcome at a time, MAJA considers four different cognitive tests and a polygenic score simultaneously. This accounts for the fact that cognitive traits are correlated and helps filter out "noise" (false positives).
  • Longitudinal Analysis: For dementia risk, they used Cox proportional hazards models linked to 17 years of Scottish healthcare records to see which protein levels at the time of blood draw predicted a future diagnosis.
  • Covariate Adjustment: The models were rigorously adjusted for age, sex, BMI, education, and even an "epigenetic smoking score" to ensure the protein signals weren't just proxies for lifestyle factors.

Datasets

  • Generation Scotland (GS): A large, family-based cohort study. The analysis included data from 14,258 individuals for cognitive traits and 14,745 for APOE status.
  • Health Record Linkage: Dementia diagnoses were obtained through linkage to primary care and secondary hospital records in Scotland.
  • External GWAS Data: A cognitive Polygenic Score (PGS) was derived from a massive meta-analysis of 263,184 individuals to represent genetic predisposition to cognitive health.

Key Insights and Ideas

  • The Selenoprotein P Connection: Higher levels of Selenoprotein P (which transports selenium to the brain) were associated with better performance in vocabulary, processing speed, and verbal fluency. This adds to a complex debate in the field, as some smaller studies previously suggested a "U-shaped" relationship where too much or too little is harmful.
  • Afamin as a Metabolic Link: The study found that carriers of the APOE ε4 allele (the strongest genetic risk factor for Alzheimer's) had significantly lower levels of Afamin. Since Afamin is involved in transporting Vitamin E and is linked to metabolic syndrome, this suggests that the APOE gene might influence dementia risk through specific metabolic pathways in the liver and blood.
  • The Immune System’s "Shield": Perhaps the most striking finding was Protein Group 21 (comprising Complement C2 and Factor B). Higher levels of these immune proteins were associated with a 25% lower hazard for incident dementia. This suggests that certain components of the innate immune system might play a protective role against neurodegeneration.

Key Results

  • Cognition: Selenoprotein P was the only protein to survive full adjustment across multiple cognitive domains (absolute $\beta$ range = 0.02 – 0.03, PIP $\ge$ 0.95).
  • Genetic Risk: APOE ε4 status was strongly associated with 11 proteomic signals. Notably, ε4 carriers showed lower Afamin ($\beta$ = -0.08 to -0.2) and, as expected, lower Apolipoprotein E levels.
  • Dementia Prediction: A higher signal for the Complement C2/Factor B group was significantly protective against dementia (Hazard Ratio = 0.75, $p < 3.1 \times 10^{-4}$). This association held true even when adjusted for APOE status and up to 17 years of follow-up.

Limitations and Future Work

  • Ancestry Bias: The study was conducted almost exclusively on individuals of European (Scottish) ancestry. The authors acknowledge that these biomarkers must be validated in more diverse global populations.
  • Causality vs. Correlation: As an observational study, it’s unclear if these proteins cause cognitive decline or are simply "smoke" from the "fire" of underlying pathology. The authors suggest using Mendelian Randomization in future studies to probe for causal links.
  • Sensitivity: While mass spectrometry is untargeted, it currently only detects the most abundant proteins in the blood. Future improvements in MS sensitivity may reveal even rarer proteins that correlate with early-stage brain changes.

Summary updated

Abstract

Untargeted mass spectrometry remains underutilised for blood-based biomarker discovery in dementia research from large cohorts, where affinity-based approaches dominate. To address this, we examined mass-spectrometry-derived proteomic correlates of cognitive function, genetic predisposition to cognitive health, <i>APOE</i> ε4 status, and incident dementia. Using multivariate Bayesian regression, we assessed associations between 439 independent mass spectrometry signals and five cognitive outcomes-four cognitive tests and a cognitive function polygenic score-in Generation Scotland (n = 14,258). We observed associations between three cognitive tests (digit symbol, vocabulary and verbal fluency) and a mass spectrometry signal that mapped to selenoprotein P (absolute β<sub>range</sub> = 0.02 - 0.03, posterior inclusion probability ≥ 0.95). Carrying one or two copies of the <i>APOE</i> ε4 allele was associated with a lower mass spectrometry signal that mapped to afamin (β = -0.08 and -0.2, respectively, p < 3.1 x 10<sup>-4</sup>, n = 14,745). A higher mass spectrometry signal that maps to both complement c2 and complement factor B proteins was associated with lower hazard for incident dementia (hazard ratio = 0.75, p < 3.1 x 10<sup>-4</sup>, n<sub>cases</sub> = 212 and n<sub>controls</sub> = 6,765) diagnosed up to 17 years after blood sampling. We identified specific independent mass spectrometry signals which may be candidate biomarkers of cognitive function, <i>APOE</i> ε4 status, and could aid in the early detection of dementia; however, further replication studies in other populations are required.
BibTeX
@misc{smith2026proteomic,
  author = {Hannah M. Smith and Joanna E. Moodie and G Davies and Anne Richmond and Janet Robertson and Aleksandra D. Chybowska and Camilla Drake and Poppy Adkin and Spyros I. Vernardis and Arturas Grauslys and Matthew E.H. White and Sergej Andrejev and Charles Brigden and Christoph B. Messner and Caroline Hayward and Aleksej Zelezniak and Markus Ralser and Ilse Krätschmer and Matthew R. Robinson and S. R. Cox and R. Marioni},
  title = {{Proteomic biomarkers of cognitive function, <i>APOE</i> ε4 status, and dementia in Generation Scotland}},
  year = {2026},
  doi = {10.64898/2026.01.27.26344912},
  howpublished = {medRxiv},
}
Figure