A discovery protein panel for brain predicted age discordance using MRI in neurologically healthy individuals

Background and Objectives

Brain age is a global measure that compares structural brain MRI with large reference datasets. Predicted age deviation (PAD) is the deviation between predicted brain age and chronological age, with positive values indicating advanced aging. Identifying blood-based biomarkers that approximate brain PAD could provide an accessible and cost-effective measure of brain health as an alternative to MRI, but no blood-based biomarkers have yet been identified. This study aimed to investigate novel blood-based biomarkers associated with accelerated PAD using an unbiased proteomics approach to discover new biomarkers.

Methods

This study is a secondary analysis with a cross-sectional case-control design using the LIMBIC-CENC dataset as a discovery approach to understand novel biomarker patterns. Brain age was estimated using brainageR in 137 participants aged ≤40 years with no substantial cognitive deficits or neurological disorders. Cases (n = 76) included individuals with brain age ≥5 years older than chronological age, whereas controls (n = 61) had brain age equal to or younger than chronological age (PAD range: -1.3 to 0; mean = -0.9) and were otherwise matched on demographics and clinical features. Unbiased proteomic profiling of ∼5,400 proteins was performed using the Olink Explore platform. Differential protein expression between groups was assessed using Wilcoxon tests with Benjamini-Hochberg correction. Receiver operating characteristic (ROC) analysis was performed on probabilities derived from generalized linear models (GLMs) to identify optimal protein combinations, prioritizing maximizing both sensitivity and negative predictive value.

Results

Olink analyses identified 418 proteins that were significantly different between groups after multiple-comparison correction. Upregulated proteins in participants with PAD≥5 years included: component inhibitor-nuclear factor kappa-b kinase (CHUK), methenyltetrahydrofolate synthetase domain containing (MTHFSD), and epidermal growth factor (EGF), with log2 fold changes of 1.70–1.80. Insulin-like peptide 3 (INSL3) was the most downregulated protein (log2 fold change −2.27). Enriched pathways involved nuclear factor kappa-b (NF-κB), heat-shock protein, and Wingless/Integrated (Wnt) signaling. Models including 6-7 dysregulated proteins (e.g., CHUK and INSL3) achieved AUCs>0.9, with sensitivities >0.90 and specificities >0.70.

Discussion

These discovery-based findings warrant validation in larger cohorts and suggest potential for blood-based protein panel detection of early, clinically silent, pre-pathological accelerated brain aging changes when interventions may be most effective.