Pooling samples prior to LC MS MS analysis has been shown to decrease intersubject variability ZD6474 and enhance the likelihood that any changes detected would be universal to disease. Prior to pooling, each control and probable AD mem brane rich protein fraction was visualized by silver stain ing following 1D gel electrophoresis to confirm equal protein contributions and to demonstrate comparable purity and integrity. Peptides were extracted from the samples following an in gel tryptic digest and analyzed in technical replicate using LC MS MS in a data dependent manner as described in the methods. After database Inhibitors,Modulators,Libraries searching, we identified and quantified 7,910 peptides representing 1,957 proteins. A total of 1,009 homologous protein groups were identified across control and AD sam ples fold difference from decreasing in AD to increasing.

Of these, 38% contain at least one transmembrane domain predicted by TMHMM 2. 0, and DAVID ontology analysis reported that 55% had Protein Information Resource annota tions relating to membrane. To determine candidate AD platelet membrane protein biomarkers from our list of 1,009 quantified proteins, we employed an approach to estimate true FDR that fully utilizes the power of technical Inhibitors,Modulators,Libraries replicates and a null experimental comparison Inhibitors,Modulators,Libraries to quantify false positives under any given filtering criteria. Relative differences in protein levels, ion intensities for identified peptides, expressed as signal to noise ratios, were extracted in MS survey scans of high resolution.

A ratio of ion intensities for the peptide Inhibitors,Modulators,Libraries precursor ions from AD and control LC MS runs were calculated, log2 transformed, and averaged to obtain a protein ratio across samples, Inhibitors,Modulators,Libraries and a null experiment log2 transformed ratio for control replicates. As predicted by the null hypothesis, the histogram of the differences and null experiment between protein log2 ratios fit Gaussian distributions, which enabled us to evaluate systematic bias according to the mean and biological var iation based on SD. The null experiment has a much smaller SD than the average log2 population. This is consistent with high reproducibility across replicates and indicates that our quantitative bioinformatics approach has suffi cient precision to detect the biological variance, which manifests as a much wider SD for the latter population.

As a filtering criterion, proteins with potentially increased or decreased abundance in AD that fell outside the 99. thorough 9% two tailed confidence interval were considered as a sub group of interest. Increased confidence in the average of two technical replicates was obtained by restricting pro teins considered significantly changed to those with a coefficient of variation of less than 100%, where this filtering criterion alone reduced false positives surviving filtering in the null experiment from 74 to 24.