Abstract
We have used proteomic fingerprinting to investigate diagnosis of Alzheimer's disease (AD). Samples of lumbar cerebrospinal fluid (CSF) from clinically-diagnosed AD cases (n = 33), age-matched controls (n = 20), and mild cognitive impairment (MCI) patients (n = 10) were used to obtain proteomic profiles, followed by bioinformatic analysis that generated a set of potential biomarkers in CSF samples that could discriminate AD cases from controls. The identity of the biomarker ions was determined using mass spectroscopy. The panel of seven peptide biomarker ions was able to discriminate AD patients from controls with a median accuracy of 95% (sensitivity 85%, specificity 97%). When this model was applied to an independent blind dataset from MCI patients, the intensity of signals was intermediate between the control and AD patients implying that these markers could potentially predict patients with early neurodegenerative disease. The panel were identified, in order of predictive ability, as SPARC-like 1 protein, fibrinogen alpha chain precursor, amyloid-β, apolipoprotein E precursor, serum albumin precursor, keratin type I cytoskeletal 9, and tetranectin. The 7 ion ANN model was further validated using an independent cohort of samples, where the model was able to classify AD cases from controls with median accuracy of 84.5% (sensitivity 93.3%, specificity 75.7%). Validation by immunoassay was performed on the top three identified markers using the discovery samples and an independent sample cohort which was from postmortem confirmed AD patients (n = 17).