Clustering biomarkers in dominantly inherited Alzheimer’s disease

Abstract

AbstractBackgroundNumerous biomarkers exist for potentially studying autosomal dominant Alzheimer disease (ADAD) pathology. Understanding the relationships between biomarkers and disease progression can inform patient care and clinical trials. We utilized machine learning‐based clustering and feature selection to identify an optimal subset of biomarkers reflective of disease stage and pathology.MethodCross sectional clinical and neuropsychological evaluation (Clinical Dementia Rating (CDR), Geriatric Depression Scale (GDS), Mini‐Mental State Exam (MMSE)), structural magnetic resonance imaging (AD cortical signature), amyloid measures (amyloid positron emission tomography (PET) and CSF Aβ42 and 40), tau measures (CSF total tau and phosphorylated tau isoforms pT181, pT205, pT202, and pT217), and neurodegeneration measures (fluorodeoxyglucose (FDG) PET, serum NFL, CSF neurogranin, SNAP‐25, VILLIP‐1) and neuroinflammation (CSF YKL‐40) were acquired in 130 mutation carriers (MC) and 79 non‐carriers (NC) from the Dominantly Inherited Alzheimer Network observational study (Table 1). Hierarchical clustering was used to group biomarkers. Decision tree‐based feature selection identified biomarkers that were the strongest predictors of mutation status and estimated years to symptom onset (EYO).ResultFigure 1 shows the absolute value correlation matrix used for hierarchical clustering. Figure 2 shows the dendrogram of the clustering results. FDG, GDS, and Aβ40 were the least similar to other biomarkers. Aβ42 and 42/40 ratio, PET amyloid, cortical signature, CDR, MMSE, and NFL were similar. With the exception of YKL‐40, the “emerging” biomarkers (neurogranin, VILLIP‐1, SNAP‐25) clustered together. Total tau and tau isoforms were grouped together. Feature selection identified YKL‐40, pT205, and pT217 as the strongest predictors of EYO in MCs (Figure 3). YKL‐40 had the highest correlation with EYO (.59,p‐value<.001), suggesting a possible role of persistent increasing neuroinflimation. With regard to mutation status, the strongest predictors were measures related to amyloid, including Aβ42/40 ratio, amyloid PET PIB, and pT217 (Figure 4).ConclusionUnderstanding the relationships biomarkers have with each other, as well as understanding which biomarkers are most reflective of disease processes, is vital for patient care and clinical trials. In this study, we have identified groups of biomarkers that are most similar to each other, as well as identified which biomarkers are most reflective of EYO and mutation status in ADAD.