Kleineidam, Luca, Demnitz‐King, Harriet, Adami, Pamela V Martino, García‐González, Pablo, Cano, Amanda, Boada, Mercè, Peters, Oliver, Priller, Josef, Schneider, Anja, Fliessbach, Klaus, Wiltfang, Jens, Bürger, Katharina, Perneczky, Robert, Teipel, Stefan, Laske, Christoph, Spottke, Annika, Heneka, Michael T., Schmid, Matthias, Düzel, Emrah, and Frikke‐Schmidt, Ruth
Background: Age is the strongest risk factor for dementia, including Alzheimer's disease (AD). However, across individuals of the same chronological age there is considerable heterogeneity in rates of cognitive decline. These differences have been attributed to individual differences in age‐related biological processes, including brain protein expression. To capture these differences, we developed a measure of protein brain age from cerebrospinal fluid (CSF)‐proteins, and validated its predictive utility. Methods: In older adults across the dementia continuum enrolled in DELCODE (training cohort; N = 488, 5‐year follow‐up) and patients with subjective cognitive decline and mild cognitive impairment (MCI) from the memory‐clinic based Fundació ACE (FACE) cohort (test cohort; N = 509, 3‐year follow‐up), 1063 CSF‐proteins were measured using OLINK‐technology. Proteins optimally predicting chronological age in A/T/N‐negative individuals without dementia from DELCODE (N = 196) were selected based on a LASSO‐regression and summarized into a BrainProteinAge‐score for all participants. A harmonized, age‐sensitive cognitive composite was created and the association between BrainProteinAge and cognitive decline assessed using structural equation modeling. Further, the predictive utility of ProteinBrainAge for incident all‐cause‐dementia was examined using time‐dependent ROC analysis in MCI patients from the test cohort. Results: 118 CSF‐proteins were selected and used to calculate BrainProteinAge. In the training (Est[SE] = 0.06[0.02],p<0.001) and test (Est[SE] = 0.04[0.02],p = 0.013) cohorts, BrainProteinAge predicted cognitive decline beyond demographic characteristics (chronological age, sex, education) and A/T/N profiles. ProteinBrainAge captured 40% of the processes linking chronological age and cognitive decline, and improved prediction to all‐cause dementia beyond demographic characteristics and A/T/N profiles in MCI‐patients (1‐year AUC: 0.78 versus 0.82, p(difference) = 0.006). Specifically, within the AD‐continuum, BrainProteinAge (HR[CI] = 1.056[1.004‐1.110]), compared to chronological age (HR[CI] = 1.022[0.990‐1.054]), has greater predictive utility. Whilst an 8.6‐year ProteinBrainAge difference corresponded to the same risk increase observed between A+/T+ and A+/T‐ MCI patients, a difference of 24.6‐years was required for chronological age. All results remained substantively unchanged when using an optimized BrainProteinAge based on 10 proteins most strongly linked to cognitive decline. Conclusions: A new measure of biological age improves individualized prediction of cognitive decline and conversion to dementia in memory clinic patients. This provides an opportunity for precision medicine and a better understanding of age‐related biological processes contributing to dementia. [ABSTRACT FROM AUTHOR]