Peak-Driven NMR Binning Combined with STOCSY Identifies Metabolic Biomarkers of Alzheimer’s Disease in Cerebrospinal Fluid

Abstract Alzheimer’s disease (AD) is marked by broad metabolic dysregulation, making cerebrospinal fluid (CSF) a direct window into brain metabolism. Very high-field 900 MHz 1H and 1H−13C HSQC NMR spectroscopy profiled CSF from 66 patients (38 AD, 28 non-AD with other cognitive disorders). High spectral resolution facilitates peak-driven binning, which centers each bin on its resonance apex to keep every variable tied to one signal. Univariate testing identified 69 discriminant 1H bins and 17 discriminant HSQC cross-peaks (p ≤ 0.10). These bins were correlated by statistical total correlation spectroscopy (STOCSY) into clusters and assigned to metabolites, yielding nine discriminant metabolites after correction for multiple testing (q ≤ 0.05). All discriminant metabolites were lower in AD, with fold changes around 0.7, spanning branched-chain and aromatic amino acids alongside energy-metabolism intermediates. An orthogonal partial least squares discriminant analysis (OPLS-DA) model built on these metabolites separated AD from this clinically heterogeneous non-AD group under leave-one-out cross-validation (AUC = 0.87, accuracy = 0.77, Q2 = 0.37, permutation p = 0.02). These findings identify a small panel of biologically interpretable metabolites with potential utility for differentiating AD from clinically similar cognitive disorders.