Comparing MRI and CSF biomarkers in Alzheimer's disease: Intergroup discrimination and predicting clinical change

Abstract

BACKGROUND AND METHODS The pathological hallmarks of Alzheimer disease (AD) are the presence of neurofibrillary tangles (NFT) composed of hyperphosphorylated tau and neuritic plaques composed of β-amyloid (Aβ) fibrils. Biomarker and imaging indicators of disease that closely reflect the underlying pathology will add great value to clinical assessment as well as to the understanding of underlying mechanisms of AD. In this study we compared two core biochemical and imaging biomarkers, CSF and structural MRI. Two plasma cerebrospinal fluid (CSF) biomarkers in AD have been found to be promising: total tau (t-tau) and Aβ1-42. High CSF t-tau protein reflects neuronal and axonal neurodegeneration and Aβ1-42  is a major component of amyloid plaques and decrease of Aβ1-42  is thought to reflect deposition of soluble Aβ in neuritic plaques. Structural MRI captures disease related structural changes in the brain by measuring brain volume loss, the direct result of loss of neurons, synapses and supporting cellular structures. There is ample evidence supporting that MRI is an approximate in-vivo indicator of neuronal pathology in AD. A technique developed in our lab condenses the degree and location of AD related atrophy on the three dimensional T1-weighted MRI scan into a single number which is called STructural Abnormality iNDex (STAND)-score and correlates well with postmortem NFT Braak stages [1, 2]. In this work, we use STAND-scores as an indicator of the severity or stage of the AD-like pattern of volume loss on structural MRI. The aims of this work were two-fold in the context of evaluating both the biomarkers: cross-sectional clinical correlations and prediction of future clinical change. In this work, we examine both the aforementioned questions using data from the Alzheimer’s disease Neuroimaging Initiative (ADNI) study which consists of large database of normal elderly (CN), amnestic mild cognitive impairment (aMCI) and AD with both MRI and CSF biomarkers. RESULTS Cross Sectional Clinical Correlations: (a) Among all subjects, the correlation between STAND and cognitive scores was stronger than between the CSF and cognitive scores (p<0.01, Choi’s test), suggesting STAND is more closely related to cognitive performance than CSF biomarkers. When the subjects were split into groups by clinical diagnosis, there was no significant correlation between the CSF biomarkers and cognitive scores within any of the individual groups. However STAND-score correlated significantly with widely used indices of general cognition the Clinical Dementia Rating sum of boxes’ (CDR-SB) and the Mini Mental State Exam (MMSE) in the aMCI and AD groups suggesting that only structural MRI is closely related to within group variation in cognitive status in aMCI and AD. (b) Each of the MRI/CSF biomarkers independently contributed (p<0.001) to the prediction of clinical group membership (in univariate models). The model that combined STAND, t-tau and Aβ1-42, had better performance that any one disease indicator alone, the contribution of each disease indicator remained significant (p<0.001) and the biggest contributor in the combined model was STAND-scores. Area under ROC (AUROC) and diagnostic