Background
Cerebrospinal fluid (CSF) is an important biofluid for biomarkers of neurodegenerative diseases such as Alzheimer's disease (AD). By employing tandem mass tag (TMT) proteomics, thousands of proteins can be quantified simultaneously in large cohorts, making it a powerful tool for biomarker discovery. However, TMT proteomics in CSF is associated with analytical challenges regarding sample preparation and data processing. In this study we address those challenges ranging from data normalization over sample preparation to sample analysis. Method: Using liquid chromatography coupled to mass-spectrometry (LC-MS), we analyzed TMT multiplex samples consisting of either identical or individual CSF samples, evaluated quantification accuracy and tested the performance of different data normalization approaches. We examined MS2 and MS3 acquisition strategies regarding accuracy of quantification and performed a comparative evaluation of filter-assisted sample preparation (FASP) and an in-solution protocol. Finally, four normalization approaches (median, quantile, Total Peptide Amount, TAMPOR) were applied to the previously published European Medical Information Framework Alzheimer's Disease Multimodal Biomarker Discovery (EMIF-AD MBD) dataset.
Conclusion
We have evaluated several analytical aspects of TMT proteomics in CSF. The results of our study provide practical guidelines to improve the accuracy of quantification and aid in the design of sample preparation and analytical protocol. The AD biomarker list extracted from the EMIF-AD cohort can provide a valuable basis for future biomarker studies and help elucidate pathogenic mechanisms in AD.
Results
The correlation of measured TMT reporter ratios with spiked-in standard peptide amounts was significantly lower for TMT multiplexes composed of individual CSF samples compared with those composed of aliquots of a single CSF pool, demonstrating that the heterogeneous CSF sample composition influences TMT quantitation. Comparison of TMT reporter normalization methods showed that the correlation could be improved by applying median- and quantile-based normalization. The slope was improved by acquiring data in MS3 mode, albeit at the expense of a 29% decrease in the number of identified proteins. FASP and in-solution sample preparation of CSF samples showed a 73% overlap in identified proteins. Finally, using optimized data normalization, we present a list of 64 biomarker candidates (clinical AD vs. controls, p < 0.01) identified in the EMIF-AD cohort.