Unbiased proteomics and multivariable regularized regression techniques identify SMOC1, NOG, APCS, and NTN1 in an Alzheimer’s disease brain proteomic signature

Abstract Advancements in omics methodologies have generated a wealth of high-dimensional Alzheimer’s disease (AD) datasets, creating significant opportunities and challenges for data interpretation. In this study, we utilized multivariable regularized regression techniques to identify a reduced set of proteins that could discriminate between AD and cognitively normal (CN) brain samples. UtilizingeNetXplorer, an R package that tests the accuracy and significance of a family of elastic net generalized linear models, we identified 4 proteins (SMOC1, NOG, APCS, NTN1) that accurately discriminated between AD (n = 31) and CN (n = 22) middle frontal gyrus (MFG) tissue samples from Religious Orders Study participants with 83 percent accuracy. We then validated this signature in MFG samples from Baltimore Longitudinal Study of Aging participants using leave-one-out logistic regression cross-validation, finding that the signature again accurately discriminated AD (n = 31) and CN (n = 19) participants with a receiver operating characteristic curve area under the curve of 0.863. These proteins were strongly correlated with the burden of neurofibrillary tangle and amyloid pathology in both study cohorts. We additionally tested whether these proteins differed between AD and CN inferior temporal gyrus (ITG) samples and blood serum samples at the time of AD diagnosis in ROS and BLSA, finding that the proteins differed between AD and CN ITG samples but not in blood serum samples. The identified proteins may provide mechanistic insights into the pathophysiology of AD, and the methods utilized in this study may serve as the basis for further work with additional high-dimensional datasets in AD.