Publication highlights update - January 2025

Breakthroughs in population-scale proteomics from 2024

In this first publications highlight blog of 2025, we look back to some of the many extraordinary studies from last year that used data from the pilot phase (~3,000 proteins measured in >54,000 samples using Olink® Explore 3072) of the UK Biobank Pharma Proteomics Project (UKB-PPP).  Made available to the wider scientific community, this dataset has resulted a plethora of impactful publications across a broad range of biological questions and disease areas. The articles highlighted here represent just a fraction of the outstanding body of work published using UKB-PPP data and you can see a more comprehensive list here.

Aging and age-related diseases are now a major focus for biomedical research and a better understanding of the underlying biology will be crucial for future healthcare. Proteomics at population scale is already having a major impact in this field and the UKB-PPP data has enabled some extraordinary insights. Two examples are described below – the first examining the proteomics of aging at the level of individual organs, while the second applies this approach to the fascinating subject of brain aging. The third example leverages the data to identify biomarkers that can predict multiple types of future dementia, with some proteins showing changes up to a decade prior to clinical diagnosis. The power of large-scale proteomics for disease risk prediction is also emphatically demonstrated in the final example, where models based on small numbers of proteins provided accurate prediction of multiple common and rare diseases.

These outstanding studies provide a glimpse of what may come from the recently announced major expansion of the project, which aims to measure over 5,400 proteins in 600,000 samples from UKB participants using Olink® Explore HT – READ MORE.

Organ-specific aging clocks and human disease.

A team from Harvard Medical School approached the issue of biological vs chronological aging from the perspective of specific organs, gaining some remarkable insights into the nature of chronic diseases. Organ-specific aging clocks were derived using proteins with at least 4-fold higher expression in the selected organ compared to all others.

Key highlights:

• Derived a general aging model with robust correlation to chronological aging and a general mortality model (hazard ratio 1.37) that indicated higher overall mortality risk in men than women.
• Identified specific clocks for 18 different organs that could predict mortality and organ-specific diseases, while identifying proteins that mediate age-related disease mechanisms.
• Organ-specific aging models suggested that chronic diseases represent system and sub-system aging that is modifiable through environmental factors.

Plasma proteomics identify biomarkers and undulating changes of brain aging

The concept of organ-specific aging was applied to a study of brain aging by a group from Fudan University, Shanghai, who used multimodal MRI and machine learning models to stratify UKB participants on the basis of their brain age gap (BAG) – the deviation between predicted and chronological age, which has been proposed as an estimate brain health status. Protein expression was assessed in relation to BAG and actual age to identify relevant new markers and insights into key pathways and how their relative importance may change dynamically with brain age.

Key highlights:

• 13 proteins were significantly associated with BAGs, with the top associated pathway related to stress, regeneration, and inflammation.
• GDF15 and BCAN showed the most significant positive and negative BAG associations respectively, and these wo markers also showed multiple disease associations (including dementia, stroke, and movement functions), while BCAN levels also showed associations with multiple brain structural measurements
• Proteogenomic analysis showed that genetically predicted BCAN is causally related to lower BAG and this protein may be a suitable target for future drug development
• Undulating changes were seen in the plasma proteome across brain aging, with brain age-related change peaks at 57, 70 and 78 years, implicating distinct biological pathways during brain aging.

Plasma proteomic profiles predict future dementia in healthy adults

The tremendous value of population-scale multiomic and clinical datasets to improve future healthcare was demonstrated in a study that looked for protein risk scores that could predict future incidence of dementia. Baseline associations of protein levels with all-cause dementia (ACD), Alzheimer’s disease (AD) and vascular dementia (VaD) were all examined with a follow-up time of over 14 years.

Key highlights:

• Multiple associations were identified including both known and novel proteins, with GDF15, NEFL, GFAP and MMP12 most consistently associated with all three forms of dementia.
• Predictive modeling combining either GFAP or GDF15 with demographic variables enabled accurate prediction of all three forms of dementia (AUCs 0.872 to 0.912).
• The GFAP/demographic model retained high predictive performance even for individuals who were not clinically diagnosed until 10 years later.
• Individuals with high baseline GFAP levels were more than 2 times more likely to develop future dementia overall.

Proteomic signatures improve risk prediction for common and rare diseases

A group from the MRC Epidemiology Unit at Cambridge University used the UKB-PPP dataset to to find “sparse protein-based predictors” of 5-20 proteins that could predict 10-year incident disease for rare and common diseases. They wanted to identify signatures based on limited numbers of proteins that could  eventually be translated into a clinical setting, and to evaluate their predictive power compared to individual and combined models incorporating clinical parameters, current clinical biomarkers, and polygenic risk scores (PRS). Remarkably, they could identify  67 sparse prediction models that provided clinically relevant improvements over current models.

Key highlights:

• As well as out-performing clinical models, the 67 multi-protein models substantially improved them in combined models (median improvement ~7%).
• Of 23 diseases with available PRSs in the UKB data, the genetic data only improved the basic clinical model in 7 cases, with a median improvement of just 3% (compared to 8% for the protein models for these 7 diseases).
• 6 of the models could be compared with Olink data from a much smaller, independent cohort and all were validated with high concordance (r=0.81 for predictive accuracy and r=0.97 for improvement of clinical models).
• While the UKB-PPP dataset was only trained for 10-year prediction, the models retained good performance for 20-year disease incidence in the validation cohort.