Brain insulin action plays an important role in metabolic and cognitive health, but there is no biomarker available to assess brain insulin resistance in humans. Here, we developed a machine learning framework based on blood DNA methylation profiles of participants who did not have type 2 diabetes with and without brain insulin resistance and detailed metabolic phenotyping. We identified 540 DNA methylation sites (CpGs) as classifiers of brain insulin resistance in a discovery cohort (n = 167), results that were validated in two replication cohorts (n = 33 and 24) with high accuracy (83 to 94%). All 540 CpGs were differentially methylated and annotated to 445 genes mapping to neuronal development and axonogenesis processes. Methylation patterns of 98 of 540 CpGs exhibited a strong and significant (P < 0.05) blood-brain correlation, indicating that blood cells are a reliable proxy to capture brain-specific DNA methylation changes. These blood-based epigenetic signatures could potentially serve in the future for the early detection of individuals with brain insulin resistance in a broad clinical setting.
GrantsGerman research Association as a clinician scientist German research Foundation German diabetes Society (ddG) German center for diabetes research Brandenburg State German Ministry of Education and Research (BMBF)