Abstract
Alzheimer's disease (AD), the most common form of dementia in the elderly, affects one in eight people by the time they are 65 years old and leads to a progressive deterioration of cognitive function. An early diagnosis of AD is essential for maximizing the potential benefit of pharmacological therapy. However, diagnosing AD requires input from both psychological and radiological investigations, and in most cases the disease has persisted for many years before its presence is formally confirmed. In order to address these diagnostic pitfalls, Ray et al. adopted a multi-analyte approach to examine whether plasma-borne signalling proteins could be used to predict clinical AD in unknown subjects.
Using an arrayed sandwich ELISA, the authors measured the abundance of 120 signalling proteins in an archived collection of 259 samples from both AD patients and non-demented controls. Samples were then divided into two main groups, consisting of a ‘training set’ for defining biochemical predictors, and a ‘test set’ for subsequently predicting blinded samples.
Using samples in the training set, a statistical algorithm named predictive analysis of microarrays (PAM) was applied to scrutinize data for the presence of an Alzheimer's specific protein expression signature. From this analysis, a bank of 18 signalling proteins (including interleukins, colony-stimulating factors and tumour necrosis factors) were identified that could be used as predictors to detect AD and classify subjects into an AD or non-AD phenotype. When subsequently applied to the test set, these 18 predictors correctly classified 90% of AD and 88% of non-AD samples, including correctly identifying 10 out of 11 subjects diagnosed with other non-AD dementias.
Further to this, Ray et al. went on to examine the ability of PAM to predict the presence of AD among a group of patients previously followed up for a longitudinal study on mild cognitive impairment (MCI). From a panel of 47 patients with a previous history of MCI, PAM correctly identified 20 out of 22 patients who progressed to AD over a 2–6 year follow-up period, while also correctly classifying all eight MCI patients who developed other dementias as non-AD subjects.
With this paper, Ray et al. raise again the potential applications of array-based technology in the diagnosis of diseases that have a complex aetiology. By identifying AD-specific alterations in the expression of 18 signalling proteins, the authors have also been able to pinpoint aberrant signalling pathways that may provide a biochemical understanding for the basis of AD. Such an approach may prove useful in the future for identifying novel pharmacological targets for the treatment of these diseases.